{"id":40103,"date":"2026-11-18T10:08:00","date_gmt":"2026-11-18T15:08:00","guid":{"rendered":"https:\/\/www.drugpatentwatch.com\/blog\/?p=40103"},"modified":"2026-11-07T22:49:35","modified_gmt":"2026-11-08T03:49:35","slug":"ai-visibility-isnt-ai-demand-pharma-is-measuring-only-half-the-problem","status":"publish","type":"post","link":"https:\/\/www.drugpatentwatch.com\/blog\/ai-visibility-isnt-ai-demand-pharma-is-measuring-only-half-the-problem\/","title":{"rendered":"AI Visibility Isn&#8217;t AI Demand: Pharma Is Measuring Only Half the Problem"},"content":{"rendered":"\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"572\" src=\"https:\/\/www.drugpatentwatch.com\/blog\/wp-content\/uploads\/2026\/11\/image-11.png\" alt=\"\" class=\"wp-image-40104\" srcset=\"https:\/\/www.drugpatentwatch.com\/blog\/wp-content\/uploads\/2026\/11\/image-11.png 1024w, https:\/\/www.drugpatentwatch.com\/blog\/wp-content\/uploads\/2026\/11\/image-11-300x168.png 300w, https:\/\/www.drugpatentwatch.com\/blog\/wp-content\/uploads\/2026\/11\/image-11-768x429.png 768w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">In its Q2 2026 pharma index, 5W Public Relations ran more than 60 patient prompts through five AI engines, five times each, and estimated that Eli Lilly received 12.5% of the citations. [5] That number describes what the engines said. It does not describe what patients asked. Those are two separate measurements, and most pharma AI-monitoring programs run only the first.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This article separates them. AI visibility is an output measure: how often a drug, brand, or company appears in AI answers. AI demand is an input measure: which questions people put to AI systems, how often, in what words, and for what purpose. A brand team can score well on the first and still be absent from the questions that drive real decisions. We show the arithmetic with public figures, define the terms, and set out what a question database adds to a monitoring program.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>The Short Answer: What Is the Difference Between AI Visibility and AI Demand?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">AI visibility is how often an AI system mentions, cites, or recommends a drug or brand in answers to a chosen set of prompts. AI demand is the volume, wording, and intent of the questions people actually ask. Visibility measures the answer. Demand measures the question.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Is AI Visibility?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">AI visibility is the share of tested AI answers in which a brand, drug, or company appears, in what position, and with what framing. It is a property of the model&#8217;s output. An analyst chooses the prompts, runs them across engines, and counts. The 5W Pharma Rx AI Visibility Index is a published example. [5]<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Is AI Demand?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">AI demand is the distribution of real questions that reach AI systems about a drug, a condition, a competitor, or a cost. It has four parts: volume, wording, intent, and context. A question such as &#8220;can I take my blood thinner with ibuprofen&#8221; has no brand name in it and still concerns a drug product. Visibility tools do not see it unless an analyst thought to write it.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>A One-Line Test for Telling Them Apart<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">If the number came from prompts your team or vendor wrote, it is visibility. If the number came from questions people chose to ask without your involvement, it is demand.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>What the Data Shows: How Many People Ask AI Systems About Health?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A large and growing share of US adults ask AI chatbots health questions. KFF found that 32% of adults did so in the past year, and OpenAI reported more than 40 million people asking ChatGPT healthcare questions each day. [1][2] These figures establish that demand exists. They do not say what the demand contains.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Has OpenAI Published About Health Questions on ChatGPT?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">OpenAI reported that more than 40 million people use ChatGPT each day for healthcare questions. [2] Healthcare accounts for over 5% of all ChatGPT messages globally. [2] About 1 in 4 of its more than 800 million regular users sends a healthcare prompt every week. [2]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Roughly 2 million messages per week concern health insurance, and seven in ten healthcare conversations take place outside normal clinic hours. [2] A separate report of the same usage data attributed roughly 600,000 of the weekly insurance messages to &#8220;hospital deserts,&#8221; defined as areas at least a 30-minute drive from the nearest hospital. [3] That report traced the figures to a post by Chengpeng Mou, OpenAI&#8217;s Head of Business Finance, not to a named OpenAI document. [3]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Analysis: using the cited figures, 600,000 of about 2 million weekly insurance messages is 30%. One quarter of 800 million regular users is about 200 million people sending a health prompt in a week. Both numbers are our arithmetic on rounded inputs. OpenAI did not report them in this form.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Does the KFF Poll Say About Who Asks and Why?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">KFF fielded its poll from February 24 to March 2, 2026, with 1,343 US adults. The margin of sampling error is plus or minus 3 percentage points. [1] Thirty-two percent of adults said they had used AI chatbots for health information in the past year. [1] Twenty-nine percent sought physical health information and 16% sought mental health information. [1]<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">32% of US adults used AI chatbots for health information in the past year (KFF poll, fielded February 24 to March 2, 2026; n=1,343; margin of error \u00b13 points). [1]<\/p>\n<\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\">Among users, 65% wanted quick information, 41% wanted to look something up before deciding whether to see a provider, and 36% preferred to look things up privately. [1] Nineteen percent cited an inability to afford a health professional. [1]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">After asking, 42% of physical-health users and 58% of mental-health users did not follow up with a clinician. [1] Forty-one percent of AI health users uploaded personal medical information, which is 13% of all adults. [1]<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>What These Two Sources Cannot Tell a Brand Team<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Neither source says which drugs people asked about, which side effects, which comparisons, or which cost questions. OpenAI holds the underlying conversations and has published aggregates only. KFF asked about topics in broad categories. A brand team that needs to know whether patients ask about its product&#8217;s dosing, its price, or a competitor&#8217;s switching protocol must obtain that information elsewhere.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>How Do AI Visibility Indices Work? A Look at the 5W Pharma Rx Index<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Visibility indices run a fixed set of prompts across several AI engines and count brand mentions or citations. The 5W index is a transparent example, which is why we use it. We are reading its published methodology, not criticizing its execution. [5]<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Did 5W Measure and How?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">5W tested more than 60 patient and consumer prompts in six sub-categories. [5] It used ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews, and ran each prompt five times per engine in clean sessions during Q2 2026. [5] 5W describes its citation shares as estimates that &#8220;should be read as directional measures of relative visibility.&#8221; [5]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Analysis: 60 prompts across 5 engines at 5 runs each gives at least 1,500 responses, since 5W says &#8220;more than 60&#8221; prompts. This is our calculation from the stated methodology. 5W did not report a response count on the page we reviewed.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Do the 5W Numbers Show When Added Up?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">5W ranked 25 companies and published shares for the top 15. [5] Lilly led at 12.5%, followed by Novo Nordisk at 11.5%, Pfizer at 8.5%, Johnson &amp; Johnson at 7.0%, and Merck at 6.0%. [5]<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Rank<\/th><th>Company<\/th><th>Estimated citation share (Q2 2026)<\/th><th>Cumulative share (our calculation)<\/th><\/tr><\/thead><tbody><tr><td>1<\/td><td>Eli Lilly<\/td><td>12.5%<\/td><td>12.5%<\/td><\/tr><tr><td>2<\/td><td>Novo Nordisk<\/td><td>11.5%<\/td><td>24.0%<\/td><\/tr><tr><td>3<\/td><td>Pfizer<\/td><td>8.5%<\/td><td>32.5%<\/td><\/tr><tr><td>4<\/td><td>Johnson &amp; Johnson<\/td><td>7.0%<\/td><td>39.5%<\/td><\/tr><tr><td>5<\/td><td>Merck<\/td><td>6.0%<\/td><td>45.5%<\/td><\/tr><tr><td>6<\/td><td>AbbVie<\/td><td>5.0%<\/td><td>50.5%<\/td><\/tr><tr><td>7<\/td><td>AstraZeneca<\/td><td>4.0%<\/td><td>54.5%<\/td><\/tr><tr><td>8<\/td><td>Moderna<\/td><td>3.5%<\/td><td>58.0%<\/td><\/tr><tr><td>9<\/td><td>Amgen<\/td><td>3.0%<\/td><td>61.0%<\/td><\/tr><tr><td>10<\/td><td>Novartis<\/td><td>2.6%<\/td><td>63.6%<\/td><\/tr><tr><td>11<\/td><td>Bristol Myers Squibb<\/td><td>2.3%<\/td><td>65.9%<\/td><\/tr><tr><td>12<\/td><td>GSK<\/td><td>2.0%<\/td><td>67.9%<\/td><\/tr><tr><td>13<\/td><td>Sanofi<\/td><td>1.8%<\/td><td>69.7%<\/td><\/tr><tr><td>14<\/td><td>Gilead Sciences<\/td><td>1.5%<\/td><td>71.2%<\/td><\/tr><tr><td>15<\/td><td>Roche<\/td><td>1.3%<\/td><td>72.5%<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Source for shares: 5W Public Relations. [5] The cumulative column is our addition.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The two top-ranked companies hold 24.0% of estimated citations between them. The top five hold 45.5%. The 15 listed companies sum to 72.5%. These are properties of the prompt set 5W chose. A different prompt set, weighted toward oncology or vaccines, would produce a different table.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>Why a Prompt Battery Reflects Its Author&#8217;s Choices<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">An analyst who writes 60 prompts decides which conditions, drugs, and phrasings count. Visibility is then a measure of the model&#8217;s behavior on that analyst&#8217;s sample of the world. This is a valid and useful measurement. It cannot also be a measurement of what patients ask, because no patient wrote those prompts.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Why Can&#8217;t Visibility Tools Measure AI Demand?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Visibility tools count outputs for prompts they supply. Demand requires counting inputs that someone else supplied. A tool that estimates demand must therefore get question data from somewhere, and the sources differ in kind. Vendors use modeled estimates, browser panels, and synthetic prompts. Platform operators hold the actual logs.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>How Do Vendors Estimate Prompt Volume Today?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt volume is a developing product category with no single method. Three published descriptions show the range.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Writesonic says its Prompt Explorer predicts monthly volume for AI prompts using two inputs: a keyword-based estimate built like Google Keyword Planner estimates, and a dataset of &#8220;2B+&#8221; AI chat conversations drawn from proprietary and licensed sources. [6] The numbers are estimates, not counts of observed prompts, and the page does not disclose the split between data sources or any independent validation. [6]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Evertune describes prompt volume as the estimated total appearances of a topic in prompts, &#8220;not unique users or queries,&#8221; and says its source is a demographically weighted panel of 25 million internet users. [8] The page does not describe how the panel data is collected, and it gives no error margins. [8]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Digiday reported in December 2025 that firms such as Profound, Semrush, and Similarweb supply prompt data to publishers. [7] Sources named include search APIs that run millions of synthetic prompts daily, clickstream data from browser extensions, and a consumer behavior panel at Semrush. [7] Digiday also reported that AI platforms do not share visibility data directly, and that the data is directional because methodologies vary. [7]<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>Synthetic Prompts<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">A synthetic prompt is one written by a person or program to resemble a real question. It measures how a model answers that wording. It does not show that anyone asked it.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>Panel and Clickstream Data<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Panel and clickstream data come from a sample of users who agreed to be observed, or whose browser activity was collected. The sample is then weighted to the general population. Accuracy depends on how well the panel represents people who ask drug questions, and the published descriptions above do not test that.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Where Do the Real Question Logs Sit?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The complete record of what people asked ChatGPT sits with OpenAI. OpenAI has published aggregates about health topics, as cited above, but not a drug-level question table. [2][3] The same is true of the other platforms. No pharma team can license a full drug-by-drug log from the platform that holds it, based on the sources reviewed here.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That is the gap. Demand data exists, but a pharma professional sees it only as vendor models, panels, or platform aggregates.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>What Does a Real Question Look Like? A Taxonomy of Pharma Question Intents<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A real question carries an intent. Visibility counts treat all mentions as equal, while questions differ in what they ask and which function owns the answer. We propose seven intent classes. This is an original framework, not an industry standard.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Original Framework: Seven Question Intents<\/strong><\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Intent class<\/th><th>Example question pattern<\/th><th>Function that owns the answer<\/th><th>Brand named in the question?<\/th><\/tr><\/thead><tbody><tr><td>Safety and side effects<\/td><td>&#8220;Is [symptom] a side effect of [drug]?&#8221;<\/td><td>Pharmacovigilance, Medical Information<\/td><td>Often<\/td><\/tr><tr><td>Dosing and administration<\/td><td>&#8220;What happens if I miss a dose of [drug]?&#8221;<\/td><td>Medical Information<\/td><td>Often<\/td><\/tr><tr><td>Interactions and contraindications<\/td><td>&#8220;Can I take [drug class] with [OTC product]?&#8221;<\/td><td>Medical Information, Regulatory<\/td><td>Sometimes<\/td><\/tr><tr><td>Access and cost<\/td><td>&#8220;Does insurance cover [drug] for [condition]?&#8221;<\/td><td>Market Access, Patient Support<\/td><td>Sometimes<\/td><\/tr><tr><td>Comparison and switching<\/td><td>&#8220;Is [drug A] better than [drug B]?&#8221;<\/td><td>Brand, Medical Affairs, CI<\/td><td>Yes<\/td><\/tr><tr><td>Off-label and unlabeled use<\/td><td>&#8220;Can [drug] be used for [unlabeled condition]?&#8221;<\/td><td>Medical Information, Regulatory, Legal<\/td><td>Often<\/td><\/tr><tr><td>Condition-level discovery<\/td><td>&#8220;What are my options for [condition]?&#8221;<\/td><td>Brand, Medical Affairs<\/td><td>No<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Source: original classification by the author, based on the functions listed in the article brief. The example patterns are illustrative.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The last row matters most for visibility. A condition-level question names no brand. Whether a brand appears in the answer depends on the model, and the brand never had a chance to appear in a prompt that mentions only the condition unless an analyst wrote that prompt.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>Questions With No Brand Name in Them<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Visibility programs usually start from brand names, because brand teams ask for brand reports. Prompts that include the brand name guarantee the brand is the topic. They say nothing about the questions where the brand might be absent and a competitor present.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Why Do Intent Classes Matter for Measurement?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Intent determines both the demand weight and the risk. A cost question and a safety question can each mention a drug once. They have different owners, different consequences when the answer is wrong, and different demand volumes. A single visibility percentage hides all of that.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>The Visibility-Demand Matrix: Four Positions a Brand Can Occupy<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Crossing the two measures yields four positions. This is an original framework. It lets a team state where a brand stands without collapsing two different numbers into one.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Covered Demand: High Demand, High Visibility<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The brand appears in answers to questions people ask often. The task here is accuracy, not presence. An answer that includes the brand but misstates a dose is a label-alignment problem.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Invisible Demand: High Demand, Low Visibility<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">People ask frequently, and the brand rarely appears. This is the position a visibility-only program cannot detect, because the questions are missing from its prompt set. A cost question in the scenario below is an example.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Unrequested Visibility: Low Demand, High Visibility<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The brand appears often in answers to questions few people ask. A high visibility score here is accurate but commercially thin. Teams that report visibility alone can present this position as success.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>Absent on Both Axes<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">The brand is neither asked about nor mentioned. For a small or new product, this is the starting point, and the plan is different from fixing a gap in a mature brand.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Position<\/th><th>What the visibility score shows<\/th><th>What demand data shows<\/th><th>First action<\/th><\/tr><\/thead><tbody><tr><td>Covered demand<\/td><td>High<\/td><td>High volume<\/td><td>Check accuracy against the label<\/td><\/tr><tr><td>Invisible demand<\/td><td>Low<\/td><td>High volume<\/td><td>Find the sources models use for these questions<\/td><\/tr><tr><td>Unrequested visibility<\/td><td>High<\/td><td>Low volume<\/td><td>Do not report as a win without demand weights<\/td><\/tr><tr><td>Absent on both<\/td><td>Low<\/td><td>Low volume<\/td><td>Decide if the question set matters to the brand<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Source: original framework by the author.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Scenario Analysis: What Happens to a Visibility Score When You Weight It by Demand?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A demand-weighted visibility score is lower than a simple average whenever visibility is highest on low-demand questions. The numbers below are a hypothetical illustration with stated assumptions. They are not measured data for any brand.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>The Formula<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Demand-weighted visibility equals the sum, across question clusters, of each cluster&#8217;s demand weight multiplied by the brand&#8217;s visibility rate in that cluster. Weights sum to 100%.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>A Worked Illustration<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Assume a hypothetical brand tested across five question clusters. The visibility rates and demand weights are invented for the illustration.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Cluster<\/th><th>Assumed demand weight<\/th><th>Assumed visibility rate<\/th><th>Weighted contribution<\/th><\/tr><\/thead><tbody><tr><td>Safety and side effects<\/td><td>30%<\/td><td>20%<\/td><td>6.00<\/td><\/tr><tr><td>Dosing and administration<\/td><td>20%<\/td><td>40%<\/td><td>8.00<\/td><\/tr><tr><td>Access and cost<\/td><td>25%<\/td><td>5%<\/td><td>1.25<\/td><\/tr><tr><td>Comparison and switching<\/td><td>15%<\/td><td>70%<\/td><td>10.50<\/td><\/tr><tr><td>Brand-name general information<\/td><td>10%<\/td><td>90%<\/td><td>9.00<\/td><\/tr><tr><td>Total<\/td><td>100%<\/td><td><\/td><td>34.75<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Source: hypothetical illustration; all inputs are assumptions.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The simple average of the five visibility rates is 45%. The demand-weighted figure is 34.75%. The gap is 10.25 percentage points. The unweighted score overstates the brand&#8217;s position because the two clusters where it appears most often hold only 25% of the assumed demand.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>What Changes if the Weights Change<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Shift the assumed weights to 25% safety, 20% dosing, 35% cost, 10% comparison, and 10% general. The weighted score falls to 30.75%. The brand&#8217;s visibility rates did not change. Only the assumption about what people ask changed. This is the reason a program needs measured weights and not assumed ones.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>What Does DrugChatter&#8217;s Question Database Add to a Monitoring Program?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">DrugChatter&#8217;s chatmon system stores pharma questions together with the answers each AI model returned. That pairing lets a team look at the question side and the answer side of the same record. We describe what the database holds and what it does not hold, because the distinction matters to anyone who cites it.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Does the chatmon Database Store?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Each record joins a question, its query type, the AI model that answered it, the answer text, and the brand project the question belongs to. Questions run on a schedule, and the weekly briefings built on this data compare results over time and across models. The questions are written in natural language, the way people ask them.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This structure supports two measurements at once. Counting questions by intent class gives a view of what is asked. Scoring the paired answers against the current FDA prescribing information gives a view of what was said. Neither requires a brand name to appear in the question.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Does the Database Not Contain?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">It is not a copy of every conversation that every consumer has had with ChatGPT, Claude, Gemini, or Perplexity. Those logs belong to the platforms. We state this plainly because a reader who assumes otherwise would misread any figure drawn from the database.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>Who Else Holds Question Data?<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">The platforms hold complete logs. Vendors including Writesonic, Evertune, Semrush, and Similarweb sell modeled or panel-based volume estimates. [6][7][8] DrugChatter&#8217;s question library is different in kind from each of these: it is pharma-specific, it is joined to full answers from several models, and it is tested repeatedly over time. We make no claim that another organization holds no pharmaceutical question data, because we cannot verify that.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Why Is a Question Library Joined to Answers Useful?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Three reasons follow from the structure. A question library lets a team weight visibility by intent class. Paired answers let a team grade each answer against the label and against safety communications. Repeated runs let a team see whether an answer changed after a label update.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">None of this replaces primary evidence. The FDA label, the clinical trial record, and the published literature remain the standard against which an AI answer is judged. The database supplies the questions and the answers to judge.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>How Accurate Are AI Answers to Real Drug Questions?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Published testing shows a meaningful rate of answers that depart from reference data. The best-known peer-reviewed example is a 2024 study of Bing Copilot in BMJ Quality &amp; Safety. It is one chatbot, tested in 2023, so it cannot describe today&#8217;s models. [4]<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What Did the BMJ Quality &amp; Safety Study Find?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Researchers asked Bing Copilot 10 questions about each of the 50 most frequently prescribed drugs in the US in 2020, which produced 500 answers. [4] They compared 484 answers with reference data. [4] Average completeness was 77%. [4] Of 484 answers, 126 (26%) did not match reference data, and 16 (just over 3%) were fully inconsistent. [4]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Seven experts reviewed a subset of 20 answers. They judged 54% aligned with scientific consensus and 39% contradictory. [4] Of those 20 answers, 22% could lead to death or severe harm if followed, according to the experts. [4] The 22% figure applies to the 20-answer subset only, not to all 500 answers. [4]<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Analysis: 126 of 484 is 26.0%, and 16 of 484 is 3.3%. These match the percentages reported in the study. The study&#8217;s DOI is 10.1136\/bmjqs-2024-017476. [4]<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Why Does Question Wording Matter to an Accuracy Finding?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The BMJ study used a fixed set of ten question types per drug. An accuracy rate from a fixed set describes performance on that set. If real questions are phrased differently, or concentrate in different intents, the rate seen by patients can differ. We do not know the direction of the difference. That is the argument for measuring the questions.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>Readability as a Separate Failure<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">The study reported an average Flesch Reading Ease score just over 37, which corresponds to degree-level difficulty. [4] An answer can match the label and still be hard for the person who asked to read. Readability is a different measure from visibility and from accuracy.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Does Regulation Already Treat Unprompted Questions Differently From Prompted Ones?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">FDA has long distinguished a question a person raises independently from information a company introduces. The agency&#8217;s draft guidance on responding to unsolicited requests for off-label information is the relevant document. It is a draft, issued in December 2011 under docket FDA-2011-D-0868, and the agency labels it as non-binding. [9]<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Why Does the Distinction Matter for AI Monitoring?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Our use of the FDA document is limited to its status and its subject. We did not extract its operative terms for this article, and a reader who needs them should read the draft and consult counsel. The relevant point is conceptual: regulators treat the origin of a question as meaningful information. A question people ask AI systems on their own is a different object from a prompt a company wrote. For medical information and regulatory teams, an off-label question that arrives unprompted raises different handling questions from a test prompt that a team wrote itself.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>What This Means for Each Function<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Each function owns a different part of the question distribution. The common requirement is a measure of what is asked, in addition to what is answered.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What This Means for Medical Affairs and Medical Information<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Medical Information teams answer questions. They already classify inquiries by type. An AI question library in the same intent classes lets them compare the inquiries they receive with the questions AI systems receive. A mismatch shows where patients and clinicians look elsewhere.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What This Means for Pharmacovigilance<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Safety questions are the intent class where a wrong answer costs the most. The KFF data shows that 42% of physical-health users and 58% of mental-health users did not follow up with a clinician. [1] A wrong answer that nobody checks does not reach a safety database. A question library does not replace case-based signal detection. It shows where answers about known labeled risks drift.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What This Means for Brand Teams<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Brand teams can use demand weights to decide which visibility gaps to close first. In the illustration above, cost and access holds 25% of assumed demand and 5% visibility. A program ranked by visibility alone would spend effort on the 90% cluster.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What This Means for Regulatory and Legal Teams<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">These teams need an evidence trail. A dated record of the question, the model, and the answer supports a review of whether a statement about the drug matched the current label. It also shows when the answer changed.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What This Means for Competitive Intelligence<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Comparison questions name a rival. A library of such questions, run across models, shows how a competitor is described next to the brand, and which sources the models draw on.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>How to Build a Demand-First AI Monitoring Program<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A demand-first program starts with the question distribution, then measures visibility and accuracy within it. The steps below are a recommended sequence, based on the framework in this article.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 1: Build the Question Set Before the Prompt Set<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Collect questions from medical information inquiries, patient support lines, call-center logs, and any vendor question data. Write each in the words people use. Tag each with an intent class from the taxonomy above.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 2: Assign Demand Weights and Record Their Source<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Weights come from measured data where available. Where they are assumptions, label them as assumptions. The scenario above shows that the assumed weights can move the result by several points.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 3: Run the Questions Across Models on a Schedule<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Record the model, date, whether web search was on, and whether citations were requested. Run each question more than once, since the same prompt can return different answers. 5W ran each prompt five times per engine for this reason. [5]<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 4: Score Visibility and Accuracy Separately<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Report the visibility rate by cluster. Report accuracy against the current prescribing information as a separate score. Do not merge them.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 5: Report the Gap<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The deliverable is the difference between the simple average and the demand-weighted figure, by cluster, with the owner of each gap named.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>What Are the Limits of This Analysis?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">This article uses public figures and one hypothetical illustration. We did not run a new test of AI models for it, and we report no new measured visibility or accuracy rates. The following limits apply.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Which Figures Are Reported and Which Are Calculated?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Figures from KFF, OpenAI, 5W, and the BMJ study are as reported by those sources or by the secondary articles cited. Cumulative shares, percentages of counts, and the response-count estimate are our arithmetic. The scenario inputs are assumptions.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What We Could Not Verify<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">We did not review the original OpenAI report document; the 40 million and 5% figures come from a trade-press summary that described the report. [2] The 600,000 and 2 million insurance figures come from a report that cites a social media post. [3] The vendor descriptions in this article reflect vendor-written pages, and none has been independently validated. [6][8]<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Key Takeaways<\/strong><\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>AI visibility measures how often a brand appears in answers to prompts an analyst chose. AI demand measures what people ask. They are separate numbers.<\/li>\n\n\n\n<li>KFF found that 32% of US adults used AI chatbots for health information in the past year, based on a poll of 1,343 adults fielded February 24 to March 2, 2026. [1]<\/li>\n\n\n\n<li>OpenAI reported more than 40 million people asking ChatGPT healthcare questions each day, and seven in ten healthcare conversations outside normal clinic hours. [2]<\/li>\n\n\n\n<li>The 5W Pharma Rx index tested more than 60 prompts across five engines at five runs each. Its 15 published company shares sum to 72.5%, by our calculation. [5]<\/li>\n\n\n\n<li>Prompt-volume vendors publish modeled or panel-based estimates. None of the vendor pages reviewed disclosed independent validation. [6][7][8]<\/li>\n\n\n\n<li>In a hypothetical illustration, demand weighting lowered a visibility score from 45% to 34.75%. Changing the weights lowered it to 30.75%.<\/li>\n\n\n\n<li>In a 2024 BMJ Quality &amp; Safety study of Bing Copilot, 26% of 484 answers did not match reference data. [4]<\/li>\n\n\n\n<li>DrugChatter&#8217;s chatmon database joins pharma questions to the answers from each AI model, which lets teams weight visibility by intent and grade answers against the label. It is not a log of every consumer conversation with an AI platform.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>FAQ<\/strong><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>What is the difference between AI visibility and AI demand in pharma?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">AI visibility is how often a drug or brand appears in AI answers to a chosen prompt set. AI demand is how often, and in what words, people ask AI systems questions about drugs, conditions, and costs. The first is an output measure and the second is an input measure.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Can a brand have high AI visibility and low demand?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Yes. If a brand appears often in answers to questions few people ask, its visibility score is high and its demand coverage is thin. We call this unrequested visibility. Weighting visibility by question volume exposes it.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>How many people ask AI chatbots health questions?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">KFF found that 32% of US adults used AI chatbots for health information in the past year. [1] OpenAI reported more than 40 million people asking ChatGPT healthcare questions each day. [2] The two figures use different populations and time frames and should not be combined.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Does AI visibility predict whether patients see a brand?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Not by itself. Visibility counts mentions in the analyst&#8217;s prompts. Whether patients see a brand depends on whether they ask the questions where it appears. 5W describes its own citation shares as directional measures of relative visibility. [5]<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Where does prompt-volume data come from?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Vendors describe modeled estimates, panels of users, browser clickstream data, and synthetic prompts. [6][7][8] Digiday reported that AI platforms do not share visibility data directly. [7] The estimates are directional and the methods vary. [7]<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Does DrugChatter hold every question people ask ChatGPT?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">No. The platforms hold their own conversation logs. DrugChatter&#8217;s chatmon database holds a library of pharma questions and the answers that each AI model returned to them, run on a schedule.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>How accurate are AI chatbots on drug information?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">It depends on the model, the date, and the questions. A 2024 BMJ Quality &amp; Safety study of Bing Copilot found that 126 of 484 answers (26%) did not match reference data. [4] That result describes one chatbot tested in 2023 and does not apply to current models.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Which function should own AI demand monitoring?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">No single function owns it. Medical Information owns dosing and interaction questions, Pharmacovigilance owns safety questions, Market Access owns cost questions, and Brand and CI own comparison questions. A shared question library with an intent tag lets each function review its own share.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Why does it matter that many health questions arrive outside clinic hours?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">OpenAI reported that seven in ten healthcare conversations take place outside normal clinic hours. [2] During those hours, an AI answer is often the only source available to the person asking. That raises the importance of label alignment in the answer.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>How often should a team re-run its question set?<\/strong><\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The sources reviewed here do not set a standard interval. A practical rule is to re-run on a schedule that catches label updates, new safety communications, and model changes, and to run each question more than once per cycle, as 5W did with five runs per engine. [5]<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>References<\/strong><\/h2>\n\n\n\n<ol class=\"wp-block-list\">\n<li>KFF. (2026, March 25; updated 2026, April 9). <em>Poll: 1 in 3 adults are turning to AI chatbots for health information, equaling the share who use social media for health.<\/em> KFF Health Information and Trust. <a href=\"https:\/\/www.kff.org\/health-information-trust\/poll-1-in-3-adults-are-turning-to-ai-chatbots-for-health-information-equaling-the-share-who-use-social-media-for-health\/\">https:\/\/www.kff.org\/health-information-trust\/poll-1-in-3-adults-are-turning-to-ai-chatbots-for-health-information-equaling-the-share-who-use-social-media-for-health\/<\/a><\/li>\n\n\n\n<li>Fierce Healthcare. (2026, January). <em>40M people use ChatGPT to get answers to healthcare questions, OpenAI says.<\/em> <a href=\"https:\/\/www.fiercehealthcare.com\/ai-and-machine-learning\/40m-people-use-chatgpt-answer-healthcare-questions-openai-says\">https:\/\/www.fiercehealthcare.com\/ai-and-machine-learning\/40m-people-use-chatgpt-answer-healthcare-questions-openai-says<\/a><\/li>\n\n\n\n<li>Bastian, M. (2026, April 6). <em>OpenAI reveals 600,000 weekly health queries from hospital deserts as seven in ten come after hours.<\/em> The Decoder. <a href=\"https:\/\/the-decoder.com\/openai-reveals-600000-weekly-health-queries-from-hospital-deserts-as-seven-in-ten-come-after-hours\/\">https:\/\/the-decoder.com\/openai-reveals-600000-weekly-health-queries-from-hospital-deserts-as-seven-in-ten-come-after-hours\/<\/a><\/li>\n\n\n\n<li>Andrikyan, W., et al. (2024). Artificial intelligence-powered chatbots in search engines: a cross-sectional study on the quality and risks of drug information for patients. <em>BMJ Quality &amp; Safety.<\/em> https:\/\/doi.org\/10.1136\/bmjqs-2024-017476. Press release: BMJ Group. (2024, October 11). <a href=\"https:\/\/bmjgroup.com\/dont-rely-on-ai-chatbots-for-accurate-safe-drug-information-patients-warned\/\">https:\/\/bmjgroup.com\/dont-rely-on-ai-chatbots-for-accurate-safe-drug-information-patients-warned\/<\/a><\/li>\n\n\n\n<li>5W Public Relations. (2026). <em>Pharma Rx AI Visibility Index 2026.<\/em> <a href=\"https:\/\/www.5wpr.com\/ai-visibility-index\/pharma-rx-ai-visibility-index-2026\/\">https:\/\/www.5wpr.com\/ai-visibility-index\/pharma-rx-ai-visibility-index-2026\/<\/a><\/li>\n\n\n\n<li>Writesonic. (2025, August 7; updated 2026, July 28). <em>Introducing Prompt Explorer.<\/em> <a href=\"https:\/\/writesonic.com\/blog\/introducing-prompt-explorer\">https:\/\/writesonic.com\/blog\/introducing-prompt-explorer<\/a><\/li>\n\n\n\n<li>Guaglione, S. (2025, December 15). Publishers are hunting for AI prompt data. Now they&#8217;re starting to get it. <em>Digiday.<\/em> <a href=\"https:\/\/digiday.com\/media\/publishers-are-hunting-for-ai-prompt-data-now-theyre-starting-to-get-it-from-tk\">https:\/\/digiday.com\/media\/publishers-are-hunting-for-ai-prompt-data-now-theyre-starting-to-get-it-from-tk<\/a><\/li>\n\n\n\n<li>Evertune. (2025, December 16). <em>How to track prompt volumes in GEO.<\/em> <a href=\"https:\/\/www.evertune.ai\/research\/insights-on-ai\/how-to-track-prompt-volumes-in-geo\">https:\/\/www.evertune.ai\/research\/insights-on-ai\/how-to-track-prompt-volumes-in-geo<\/a><\/li>\n\n\n\n<li>U.S. Food and Drug Administration. (2011, December). <em>Responding to unsolicited requests for off-label information about prescription drugs and medical devices: Draft guidance for industry<\/em> (Docket FDA-2011-D-0868). <a href=\"https:\/\/www.fda.gov\/regulatory-information\/search-fda-guidance-documents\/responding-unsolicited-requests-label-information-about-prescription-drugs-and-medical-devices\">https:\/\/www.fda.gov\/regulatory-information\/search-fda-guidance-documents\/responding-unsolicited-requests-label-information-about-prescription-drugs-and-medical-devices<\/a><\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>In its Q2 2026 pharma index, 5W Public Relations ran more than 60 patient prompts through five AI engines, five [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":40104,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_lmt_disableupdate":"","_lmt_disable":"","site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[10],"tags":[],"class_list":["post-40103","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-insights"],"modified_by":"DrugPatentWatch","_links":{"self":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts\/40103","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/comments?post=40103"}],"version-history":[{"count":1,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts\/40103\/revisions"}],"predecessor-version":[{"id":40105,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/posts\/40103\/revisions\/40105"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/media\/40104"}],"wp:attachment":[{"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/media?parent=40103"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/categories?post=40103"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.drugpatentwatch.com\/blog\/wp-json\/wp\/v2\/tags?post=40103"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}