1. TL;DR

Brands started paying for AI visibility scores in 2026, and most of those scores measure less than they claim. A tool runs a prompt on an engine, checks whether your name appears, and sells the result as a number. The number moves next week for reasons nobody logs. Real measurement is less glamorous: many prompts, repeated runs, several engines, a record of who took your place, and a check on what the answer said about you. The gap between the two is now documented. Google's top-10 pages appear in AI answers less than half the time, and the overlap with ChatGPT sits near 2% (Semrush, March 2026). The same prompt returns the same brand list less than 1 time in 100 (SparkToro, 2025). This report walks twelve tracking failures, and each one ends with a check you can run against your own setup today.

THREE TAKEAWAYS
- A score built on one run is noise. Repeated runs of an identical prompt return a matching brand list under 1% of the time (SparkToro, 2025), and the IAB's measurement standard rates anything under 50 queries as exploratory (IAB, August 2026). Ask any vendor for runs per prompt per week before you trust the number.
- Google rank and AI presence are separate games. Pages in Google's top 10 appeared in at least one AI answer 44.3% of the time, and in ChatGPT 2.1% of the time (Semrush, 2026). Top-10 pages fell from 76% to 38% of AI Overview citations in under a year (Ahrefs, 2026).
- In wellness, wrong beats absent as the state that costs you. The FDA issued its first warning letter naming inappropriate AI use in April 2026 (FDA via ECA Academy, 2026), and clinical benchmarks show AI answers fabricate medical references at high rates. A tracker that counts mentions and skips accuracy misses the risk that matters in a regulated category.

2. The Twelve Issues

1. One prompt, one engine, one day, sold as a score
2. Every tool measures a different thing
3. Ranking first on Google and appearing in AI answers are separate jobs
4. Your JavaScript hides the site from AI crawlers
5. Brand-name tracking misses the prompts that decide the sale
6. Single-engine coverage
7. Mentions counted, displacement ignored
8. Snapshots with no trend behind them
9. The visibility score rises while the clicks fall
10. Citations counted as wins, accuracy left unchecked
11. Nobody inside the company owns the number
12. Generic advice applied to a regulated category

01. One prompt, one engine, one day, sold as a score

What goes wrong. A tool runs a handful of prompts, records whether your name appeared, and converts the result into a single figure. The figure changes next week, and nobody can say why.

Why it happens. AI answers vary between runs of the same prompt. SparkToro and Gumshoe.ai ran 2,961 prompt executions across ChatGPT, Claude, and Google's AI in late 2025 and found under a 1-in-100 chance that ChatGPT or Google returns the same brand list twice, even though established brands still surfaced in 55 to 77% of runs (SparkToro via Search Engine Journal, 2025). The signal lives in the mention rate across runs. A single run samples the noise.

How to check. Ask your vendor for the prompt count, the runs per prompt, and the schedule. The IAB's measurement standard treats programs under 50 queries as exploratory, a tier below directional (IAB, 2026). A dashboard selling exploratory data as a decision-grade score has told you what it is.

What to do. Report mention rates with a spread, on a fixed panel, run on a schedule. Academic work on AI search measurement puts the floor near seven runs per prompt before the error settles (arXiv, 2026).

02. Every tool measures a different thing

What goes wrong. Two tools report two different visibility scores for the same brand in the same week, and both look authoritative.

Why it happens. More than 20 companies now sell AI visibility measurement, each with its own methodology, and the same brand can score differently on every one of them (IAB, 2026). One vendor counts mentions, another counts citations to your domain, a third weights position inside the answer. The two even diverge in the data: in Semrush's category analysis, the most-cited domain matched the most-mentioned brand only 21% of the time (Semrush via Search Engine Land, 2026). Standards arrived late. The IAB published "Measuring Visibility in the AI Era" in August 2026, defining presence, prominence, portrayal, and persuasion, plus a line between directional and decision-grade data. AMEC issued its own measurement principles in May 2026. Meanwhile 16% of brands run structured tracking of AI search performance at all (McKinsey, 2025).

How to check. Read the methodology page. Ask what one point of the score means. A vendor that will not publish what it counts is asking for trust it has not earned.

What to do. Pick one definition and hold it. Track four states across a fixed prompt set: named, cited, absent, or wrong. Consistency inside your own data beats comparison across vendors.

03. Ranking first on Google and appearing in AI answers are separate jobs

What goes wrong. A brand holds position one for its main term, assumes AI answers follow, and finds competitors named instead.

Why it happens. The overlap is small and shrinking. Semrush compared Google's top-10 pages against the pages cited by four AI platforms across ten SaaS queries in February 2026: 44.3% appeared in at least one AI answer, and the ChatGPT overlap sat at 2.1% (Semrush, 2026). The sample was small and the data was Semrush's own, and the direction matches larger independent work: Ahrefs tracked 863,000 keywords and watched top-10 pages fall from 76% of AI Overview citations in July 2025 to 38% by early 2026 (Ahrefs via Search Engine Journal, 2026), and across 15,000 long-tail queries, only about 12% of AI-cited URLs ranked in Google's top 10 for the original prompt (Ahrefs, 2025). Engines build answers from earned media, communities, and reference sites, and a brand's own site supplies 5 to 10% of what they reference (McKinsey, 2025).

How to check. Take your top five ranking pages. Ask each engine the question that page targets. Note whether your brand appears at all.

What to do. Treat AI presence as its own program. Rankings help, and the sources carry the answer.

04. Your JavaScript hides the site from AI crawlers

What goes wrong. The site renders for humans and ranks on Google, and AI engines describe your category as if you never published a word.

Why it happens. GPTBot, ClaudeBot, and PerplexityBot fetch raw HTML without executing JavaScript. An analysis of more than 500 million GPTBot fetches on Vercel infrastructure found zero evidence of JavaScript execution (Lantern, 2026). Google's crawler renders JavaScript, so Gemini and AI Overviews see the full page while every other engine sees the shell. A client-side-rendered product page can rank and stay invisible to most AI answers at the same time.

How to check. Open your top pages and view the page source, or load them with JavaScript disabled. Search the raw HTML for your key claims and product copy. Missing there means missing to the crawlers.

What to do. Server-side render the pages that answer buyer questions, starting with product, comparison, and FAQ pages.

05. Brand-name tracking misses the prompts that decide the sale

What goes wrong. A dashboard tracks "[your brand]" and reports healthy visibility. Meanwhile the category prompt, the one a buyer types, never returns your name.

Why it happens. Searching your own name is a confirmation query. Buyers ask category, comparison, trust, and timing questions first, and engines answer those with recommendations: in BrightEdge's automotive sample, AI named brands in about 97% of answers to non-branded prompts (BrightEdge, 2026, vendor data). A brand can win every branded prompt and sit absent from the answers that build the shortlist.

How to check. Split your prompt list into brand prompts and category prompts. Count each. A list weighted toward your own name flatters you.

What to do. Build the panel from buyer questions across four stages: discovery, comparison, trust, and purchase timing. Your brand belongs in the trust and timing slots. Your rivals belong in the comparisons.

06. Single-engine coverage

What goes wrong. A brand tracks ChatGPT, reads a stable number, and stays absent from Perplexity and Google AI Overviews without knowing it.

Why it happens. The engines barely read the same web. Overlap between ChatGPT's and Perplexity's top cited domains sits near 11% (5W, 2026). Reddit alone shows the split: above 5% of ChatGPT citations, about 24% of Perplexity citations, and 0.1% on Gemini in January 2026 (Tinuiti via MediaPost, 2026). The audiences differ too: ChatGPT passed 900 million weekly users in early 2026, Google's Gemini app reached about 950 million monthly users by mid-2026, and AI Overviews touch more than 2 billion people a month.

How to check. Run the same five prompts in ChatGPT, Perplexity, and Google with AI Overviews on. Compare.

What to do. Track the engines your buyers use, and read each one on its own. An average across engines hides the gap.

07. Mentions counted, displacement ignored

What goes wrong. A report says your brand appeared in 40% of tracked answers. The other 60% stays blank, and nobody records who filled it.

Why it happens. Counting presence is simple to build. Recording the competitor named in your place takes parsing the whole answer, and the parsing matters because AI answers concentrate hard: the top three brands hold 82.9% of AI visibility in news and media and 76.9% in consumer electronics across Semrush's 126-million-prompt index (Semrush, 2026, vendor data), category leaders keep the top spot in 90.4% of month-over-month checks (Semrush via Search Engine Land, 2026), and four supplement brands hold more than 47% of observed citation share (5W, 2026).

How to check. Open your last report. Look for competitor names. A report with none counts you and ignores the market.

What to do. Log the brand named in every answer where you are absent. That list shows which rivals own which prompts, and it points at the sources those answers drew from.

08. Snapshots with no trend behind them

What goes wrong. A brand runs a scan, sees a number, and files it. Six months later nobody can say whether visibility rose or fell.

Why it happens. Citation patterns move within weeks, and 2026 supplied the proof. Reddit collapsed from about 60% of ChatGPT citations to about 10% in mid-September 2025 before stabilizing (5W, 2026). Reddit sued Perplexity on October 22, 2025 (Reuters, 2025), and Reddit's share of Perplexity citations fell from 25% in February 2026 to 7% by April (Tinuiti via MediaPost, 2026). Google made Gemini 3 the default AI Overviews model in late January 2026, and citation patterns shifted with it (Ahrefs via Search Engine Journal, 2026). A brand leaning on one source loses ground with no warning.

How to check. Find the date on your last scan. Find the one before it. A missing second scan means you hold a snapshot.

What to do. Rerun the same prompt set on a schedule and date every log. Movement is the signal worth reading.

09. The visibility score rises while the clicks fall

What goes wrong. The dashboard trends up, organic traffic trends down, and the two reports never meet in the same meeting.

Why it happens. AI answers absorb clicks: on queries that trigger an AI Overview, organic click-through fell 61%, from 1.76% to 0.61%, across 3,119 informational queries in Seer Interactive's study, while pages cited inside the Overview earned 35% higher click-through than uncited ones (Seer via Search Engine Land, 2025). The clicks that survive carry intent: AI-referred retail traffic converted 54% better than other traffic in May 2026 (Adobe Analytics, 2026). A score that counts neither citations nor outcomes decorates a slide.

How to check. Put the visibility score, organic clicks, and AI-referred conversions for the same period on one page. Read them together.

What to do. Track cited against merely mentioned, and tie the log to referral quality in analytics. Visibility that produces cited placements and converting referrals is the version worth paying for.

10. Citations counted as wins, accuracy left unchecked

What goes wrong. An engine names your brand, the tracker records a win, and the answer contains a discontinued product, a stale price, or a health claim you never made.

Why it happens. Presence is easy to detect. Accuracy takes reading the sentence, and the error rates justify the reading: one synthesis of clinical AI benchmarks found models propagate planted medical errors in a large share of cases and fabricate authors, dates, or DOIs in more than 45% of generated medical references (Presenc, 2026). Buyers already behave as if they know: 86% of AI shoppers verify the recommendation somewhere else before buying (Product.ai, 2026).

How to check. Read the full text of five answers that mention you. Check each factual claim against your own site.

What to do. For a health or wellness brand this issue outranks the rest. An inaccurate claim inside an AI answer misleads a buyer and invites scrutiny. Fix errors at the source: the third-party page, the review listing, or the structured data the engine drew from.

11. Nobody inside the company owns the number

What goes wrong. AI visibility sits between SEO, PR, and content. Each team assumes another watches it, and the report circulates without an owner.

Why it happens. The money moved before the ownership settled. In Conductor's survey of more than 250 senior executives, 94% plan to increase AI search investment in 2026, and enterprises put about 12% of digital marketing budgets there in 2025 (Conductor via Business Wire, 2026, vendor survey). The measurement lags the spend: 45% of marketing leaders say they cannot measure their brand's AI visibility with accuracy, and 9% have the tools to track what they need (Semrush, 2026).

How to check. Name the person who reads the scan each month. Hesitation answers the question.

What to do. Give one person the prompt list, the log, and the calendar. Ownership of the number matters more than which department holds it.

12. Generic advice applied to a regulated category

What goes wrong. A wellness brand follows AI visibility advice written for software, publishes aggressive claims, and creates a compliance problem while visibility stays flat.

Why it happens. Health runs on different rules and different sources. Engines route health answers through a trust hierarchy: Healthline, Cleveland Clinic, and Mayo Clinic lead AI Overview share of voice in health care (Conductor, 2026), with government and clinical sources close behind. And the regulators moved from guidance to enforcement: the FDA issued its first warning letter naming inappropriate AI use on April 2, 2026, to a wellness-adjacent manufacturer that let AI agents draft compliance documents without review (FDA via ECA Academy, 2026). The heading was new. The principle was old: the company owns the output, whoever drafted it. The FTC applies the same logic to advertising through Operation AI Comply, its enforcement sweep against AI-assisted deception (FTC, 2024). A structure or function claim like "supports joint health" lives on one side of the line, and a disease claim lives on the other, and an AI answer repeating the wrong one about your product is still your problem.

How to check. Read the last three AI visibility articles you saved and count the wellness examples. Then read what the engines say about your product and screen every claim against the line you hold your own copy to.

What to do. Weight your effort toward the sources your category trusts: clinical and research references, credible health media, practitioner voices, and review platforms. Four supplement brands holding 47% of citation share (5W, 2026) shows how far that earned layer carries.

3. What to Do Next

Three moves cover the ground, and each takes a morning.

Write your prompt list before you buy anything. Fifteen prompts across discovery, comparison, trust, and timing tell you more about your position than any vendor demo, and the list becomes the yardstick you hold every tool against. The full method, with templates and a worked example, lives in our AI Search & GEO Report for Health and Wellness Brands. By the IAB's standard, fifteen prompts is exploratory measurement, and that is the honest tier for a founder's first pass: you run it yourself, it costs an hour, and it tells you where to look harder.

Run it across three engines, more than once, and log four states. Named, cited, absent, or wrong. Record the competitor named where you are absent, and record the claim where the answer gets you wrong.

Date the log and rerun in 30 days. One run is a snapshot. Two runs are a trend, and the trend tells you whether the work moves anything.

Prominently runs this as a program: daily prompt tracking across ChatGPT, Perplexity, and Google AI Overviews, with the four states logged per prompt and the fix attached to every alert. Run your own scan on prominently.ai to see where your brand stands today.

4. Sources

1. Semrush, "AI visibility: What it is and how to grow yours in 2026." Published March 27, 2026.
2. SparkToro and Gumshoe.ai via Search Engine Journal, "AI Recommendations Change With Nearly Every Query." Published 2025.
3. arXiv, "Don't Measure Once: Measuring Visibility in AI Search." Published 2026.
4. IAB, "Measuring Visibility in the AI Era." Published August 3, 2026.
5. AdExchanger, "IAB's New Advice On How To Measure AI Search Visibility." Published August 3, 2026.
6. AMEC via Paine Publishing, "I've Been Measuring AI Visibility Wrong. So Has Most of the Industry." Published May 2026.
7. McKinsey & Company, "New front door to the internet: Winning in the age of AI search." Published October 16, 2025.
8. Semrush via Search Engine Land, "ChatGPT topic ownership is rare." Published 2026.
9. Ahrefs via Search Engine Journal, "Google AI Overview Citations From Top-Ranking Pages Drop Sharply." Published 2026.
10. Ahrefs, "Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt." Published 2025.
11. Lantern, "AI Crawlers Do Not Render JavaScript." Published 2026.
12. BrightEdge, "AI, Auto, and the Purchase Journey: Branded vs Non-Branded Prompts." Published 2026.
13. 5W Public Relations, "The State of AI Citations 2026." Published June 2, 2026.
14. Tinuiti via MediaPost, "Reddit Emerges As Highly Cited Source In AI Engine Citations." Published February 26, 2026.
15. Reuters, "Reddit sues Perplexity for scraping data to train AI system." Published October 22, 2025.
16. Tinuiti via MediaPost, "Google AI Cites More Social, As Perplexity Pulls Back From Reddit." Published May 27, 2026.
17. Semrush, "2026 AI Visibility Index, 126 Million AI Search Prompts." Published 2026.
18. 5W Public Relations, "Supplements AI Visibility Index 2026." Published 2026.
19. Seer Interactive via Search Engine Land, "Google AI Overviews drive drop in organic, paid CTR." Published 2025.
20. Adobe Business, "AI traffic grows but retail sites lag in AI search visibility." Published 2026.
21. Presenc AI, "Medical AI Hallucination Rates 2026," a synthesis of clinical benchmarks. Published 2026.
22. Product.ai, "The 2026 Trust in AI Commerce Report." Fielded April 2026.
23. Conductor via Business Wire, "2026 AEO/GEO CMO Investment Report." Published January 13, 2026.
24. Conductor, "Health Care AEO/GEO Benchmarks." Published 2026.
25. FTC, "FTC Announces Crackdown on Deceptive AI Claims and Schemes" (Operation AI Comply). Published September 2024.
26. FDA warning letter to Purolea Cosmetics Lab via ECA Academy, "Use of AI Agents leads to the first FDA Warning Letter relating to AI." Letter dated April 2, 2026.

---

Note on source quality: Several figures come from vendor research produced by companies that sell measurement or visibility services (Semrush, BrightEdge, Conductor, 5W, Presenc, Lantern), and each is attributed by name and labeled in text where the distinction matters. Independent and primary sources anchor the core claims: Ahrefs, SparkToro, Seer Interactive, Tinuiti via MediaPost, Adobe Analytics, McKinsey, the IAB, AMEC, Reuters, the FDA, and the FTC. AI citation patterns move within weeks, so every figure carries its measurement date, and the checks in each section let you test the claims against your own brand rather than taking any score on faith.