AI visibility claims, checked against their sources
We traced widely repeated AI-visibility statistics to their original sources on 2026-10-05. Each row shows the claim as published, where it traces to, and what the source actually supports.
Summary, 22 claims
- SOURCED 5
- DISTORTED 7
- CIRCULAR 1
- UNSOURCED 8
- CONTRADICTED 1
The claims
| Claim | As published at | Traces to | Grade | What the source supports |
|---|---|---|---|---|
| “convert at 4.4 times the rate of standard organic traffic” | Emarketed, 2026-04-01 | Semrush, 2025-07-21 | CIRCULAR | Emarketed cites Lantern, which calls 4.4x a category average; the same figure first appears at Semrush. Two vendors restate one number. |
| “converts 4.4x to 23x higher than the average traditional organic visitor” | Demand Local, 2026-04-30 | Ahrefs, 2025-06-16 | DISTORTED | The 23x figure is Ahrefs' own signup data for one site, where AI search is 0.5% of traffic. It is not an industry average. |
| “in Seer Interactive's multi-vertical case study” | Marshal.ing (now machinelike.ai), 2026-04-13 | Seer Interactive, 2025-06-03 | DISTORTED | The post says multi-vertical. Seer's title says one site, with about 11,000 AI sessions against almost 14 million organic. |
| “45% of consumers had asked a generative AI tool for a local business recommendation” | Cheers, 2026-03-11 (updated 2026-09-02) | BrightLocal, 2026-03-10 | SOURCED | The figure is in the BrightLocal survey of 1,002 US adults. It means asked at least once in 12 months, not routine use. |
| “83% of restaurants completely invisible on ChatGPT, against 14% invisible on Google” | Cheers, 2026-03-11 | Local Falcon | SOURCED | Local Falcon reports 83% across 189,905 ChatGPT results, restaurant queries only, run by the vendor. Cheers says March; the primary says February 2026. |
| “63% of active AI users trust the recommendations they get” | Cheers, 2026-03-11 | BrightLocal, 2026-03-10 | SOURCED | BrightLocal reports 63% among active AI users. The base is 455 AI users, not all consumers. |
| “24% of all citations in January 2026 came from Reddit alone” | Medium (tentenco), 2026-05-17 | Tinuiti via MediaPost, 2026-02-26 | SOURCED | Tinuiti reports about 24% for January 2026 across nine product categories. Local buyer questions may differ. |
| “ChatGPT cites Reddit in more than 5% of responses” | Medium (tentenco), 2026-05-17 | Tinuiti via MediaPost, 2026-02-26 | DISTORTED | Tinuiti's 5% is Reddit's share of citations, not of responses. Semrush measured a different thing: close to 60% of responses in early August, around 10% by mid-September. |
| “Reddit accounts for 46.5% of Perplexity's citations” | Frase, 2026-03-03 (updated 2026-08-11) | no source found | UNSOURCED | The page names no study. Tinuiti's January 2026 figure for the same engine is about 24%. |
| “44.2% of all LLM citations come from the first 30% of text” | Frase, 2026-03-03 | Search Engine Land on Kevin Indig, 2026-02-18 | DISTORTED | The study covers ChatGPT only, with 18,012 verified citations. The claim widens it to all LLMs. |
| “Gartner forecast a 25% drop in traditional search engine volume” | BrandGEO (undated; read 2026-10-05) | Gartner release, 2024-02-19 (release text, no link kept) | CONTRADICTED | This page is careful to call 2026 an endpoint, but the bare claim spreads wider. Gartner issued a forecast; Wix reports Google visits up 2.4% year over year in Q1 2026. Only Q1 2026 is measured so far. |
| “88% of local businesses have no active strategy to appear in AI search results” | GrowthPro AI (updated 2026-07-11) | no source found | UNSOURCED | No source is named. The headline says invisible while the body says no strategy; those are different measures. |
| “More than 35% of consumers have used an AI tool to find a local business” | GrowthPro AI (updated 2026-07-11) | no source found | UNSOURCED | No source or method is shown. BrightLocal measured 45% for a similar behavior. |
| “AI search mention rate grew from 40% to 80%.” | GrowthPro AI (undated; read 2026-10-05) | no source found | UNSOURCED | The page gives no client, time window, query set or run count. |
| “66% of 18-24 year olds use ChatGPT to find information” | Noble, 2026-03-04 | Fractl, 2025-07-08 | SOURCED | The figure is in Fractl's 2025 survey. It covers general information seeking, not local buying. |
| “37% of consumers now begin their searches with AI tools” | Instant Press (undated) | Eight Oh Two, 2026-02-04 | DISTORTED | The survey base was 500 active AI users, not consumers in general. |
| “AI-referred sessions to websites increased 527% between January and May 2025” | Big Eye Agency, 2026-02-04 | Previsible via Search Engine Land, 2025-08-05 | DISTORTED | The data covers 19 GA4 properties and rose from 17,076 to 107,100 sessions. A small base, so the percentage is large. |
| “AI adoption for search jumped from 14% to 29.2% in six months.” | Big Eye Agency, 2026-02-04 | no source found | UNSOURCED | No citation appears in the page text. |
| “19% of customers using ChatGPT during purchasing decisions” | Master of Code, updated 2026-08-03 | First Page Sage, 2026-05-21 (cited by the page) | UNSOURCED | The cited source shows Education at 60.5% for any generative AI, from a survey of 3,161 people. The 19% was not found there. |
| “15% of all inbound sales calls directly from ChatGPT leads” | ROI Amplified, 2025-08-26 | no source found | UNSOURCED | The client is unnamed and the page shows no call-tracking method. It was published in 2025. |
| “a #1 spot in both Google organic and ChatGPT for 'Leadfeeder alternatives.'” | Minuttia, 2026-05-30 | no source found | UNSOURCED | No evidence is given for the claim. SparkToro (2026-01-28) found AI answer lists rarely repeat in the same order, so a rank in one answer is not a stable fact. |
| “30 to 70 percent fewer clicks across multiple client accounts” | Demand Local, 2026-04-30 | Pew Research, 2025-07-22 | DISTORTED | Pew (March 2025, 900 adults) found clicks in 8% of visits with an AI summary versus 15% without, about a 47% relative gap. The 30 to 70 range rests on named-site anecdotes. |
What our own record can test
NameBeam holds daily AI answers from four engines in 27 markets since 2026-08-02. For the rows below, this is the test we plan to run. No results have been computed, so none appear here. Rows with no test listed need data we do not hold, such as visitor conversions or survey responses.
- Row 5 and row 12, invisible businesses. For each market and category, take the businesses in the Google top 10 results. Count the share never named by any AI engine across all days since 2026-08-02. Report API answers and the logged-out ChatGPT app separately. Row 12 uses the same test across all business types in our 27 markets.
- Rows 7 and 9, Reddit share of Perplexity citations. Measure the share of Perplexity API citations that are reddit.com, per day and per market. Local buyer questions may differ from the product categories in the published figures.
- Row 8, Reddit in ChatGPT answers. Two measures: the share of OpenAI answers containing any reddit.com citation, and reddit.com's share of all citations. Plot both daily to see whether swings like Semrush's 60% to 10% appear.
- Row 14, run-to-run noise. For a fixed business and question, compute the standard deviation of its daily mention rate over 60 days. A claimed lift such as 40% to 80% can then be compared with that noise. We will also test whether visibility percentage over repeated runs is stable, using a split-half correlation.
- Row 21, rank consistency. Three measures, run per engine and separately for API and app: the share of days the same business is named twice in a row for the same market and question; the share of consecutive-day pairs with an identical ordered list; the share of days a business holds position 1.
- Row 22, AI Overviews. Our Google results capture can count the share of local buyer questions that show an AI Overview. It holds no click data, so it can test prevalence only, not click loss.
Method
Date checked: 2026-10-05. For each claim we read the page where it was published, followed its citation to the primary source, and compared the two. Quotes are copied from the pages as read on that date. Social posts and forum threads were not read in full and are excluded.
- SOURCED
- The number exists at the primary source and says what is claimed. We did not audit the primary source's own method.
- DISTORTED
- The number exists at the primary source, but the claim changes its scope, base, unit or meaning.
- CIRCULAR
- The sources cited restate each other, and the figure traces back to a single origin.
- UNSOURCED
- We found no primary source for the claim, or the source named does not contain it.
- CONTRADICTED
- A dated, measured source reports something different from the claim as commonly stated.
A grade describes the source chain, not the business making the claim.
Found an error? Write to [email protected] and we correct it with a dated note.