What we find when we audit AI answers

Everything here comes out of an audit we actually ran. The numbers are dated, the method is published, and the prompts are listed so you can re-run them yourself.

How often does Google show an AI Overview? The published figures, with what each measured

Fourteen pages that say what they counted when they print how often Google shows an AI Overview, and seven that print a figure without a unit, relay one, or give none, read on 25 September 2026. Each with its sample, query type, country, device and window, because a share of keywords, a share of prompts and a share of searches people made are three different things, and the post ranks none of them. Pew's 18%, March 2025, is historical.

18%

The share of 68,879 unique Google searches, made by 900 US adults in Pew Research Center's browsing panel from 1 to 31 March 2025, that produced an AI summary, in Pew's analysis of 22 July 2025, historical. Of the pages we read on 25 September 2026, it is the only figure whose page says it counted searches people actually made. Every other figure in the tables either counts a list of keywords or prompts its publisher chose or does not say what it counted, and none converts into this one.

Do AI Overviews cut clicks? The published figures, with what each one measured

Eleven studies and reports that print a figure for clicks on Google searches with an AI Overview, and Google's statements, which print none, read on 25 September 2026: Ahrefs, Seer Interactive, Amsive, Pew, BrightEdge, and two figures relayed by Press Gazette. Each with its unit, sample, country and window, because the units differ and the post ranks none of them.

23.1%

The median fall in click-through rate for the domains among the 1,000 with the most clicks in France that were most exposed to AI Overviews, those with an Overview on more than 20% of their queries, in the nine days after Google launched Overviews in France on 22 July 2026 against the 28 days before, from the Search Console data of 963 domains, in Ahrefs' study of 16 September 2026. Ahrefs puts the effect at 24.8% once a control group of barely exposed domains is taken into account. It is a change in a domain's click-through rate over all its queries, not the rate of the first result, and it does not convert into any other figure in the table.

How much traffic does AI send? Published figures, with what each measured

Published figures on the traffic AI assistants send to websites, from the pages named in the post, read on 25 September 2026, each with its unit, sample and window: share of all traffic, share of AI referrals, growth, session counts and click rates, which are different things and are not ranked against each other.

1.08%

The share of all website traffic that came from AI referrals across 1,215 enterprise customer domains in ten industries in the U.S., averaged monthly over May to September 2025, in Conductor's 2026 AEO / GEO Benchmarks Report, first published 12 November 2025 per its structured data and shown as last updated 6 July 2026; historical, because the data is from 2025. The other two studies we read that print AI's share of all traffic give 0.18% of total sessions in 2025, in an analysis of 2.3 billion sessions from January 2024 to December 2025 (WebFX, 18 March 2026, historical), and 0.5% of Ahrefs' own traffic for the 30 days the post of 16 June 2025 calls "the last 30 days" (historical). Each is a different set of sites.

Query fan-out: what Google says it is, and how many AI visibility tools now sell it

What Google and Search Engine Journal say query fan-out is, what Bing shows instead, with dates, and a count re-derived on 25 September 2026: 10 of the 19 AI visibility tools we track offer a fan-out feature on their own product, pricing, help or docs pages, each listed with the page and the sentence.

10 of 19

AI visibility tools, of the 19 we track, that offer a query fan-out feature on a product, pricing, help, docs or changelog page of their own, read on 25 September 2026 with navigation, header and footer removed and blog posts, glossaries and "coming soon" excluded. A raw search of the same pages finds the word on pages of 17 of the 19.

Where does "ChatGPT search volume" come from? What five vendors say about their prompt data

Where five AI-visibility vendors say their prompt-volume figures come from, in their own words and read on 25 September 2026: Profound's licensed panel conversations, Semrush's clickstream and keyword data, Ahrefs' keyword database, AthenaHQ's estimation model and keyword-volume provider, and Conductor's case against the metric. Each quoted and dated; none ranked, and no database size compared across vendors.

100

Searches a month: the value AthenaHQ's API reference, undated and read 25 September 2026, says a prompt's "Estimated monthly search volume" is returned as when it falls below the "~100/mo" floor that its keyword-volume provider measures.

How to see ChatGPT and Perplexity visits in your analytics

Where ChatGPT and Perplexity visits land in Google Analytics 4, what tag OpenAI says ChatGPT adds to its links, what tag its links carried in our research run of 4 September 2026, and what Perplexity's own pages say, which on the pages we read on 25 September 2026 is nothing either way.

151 of 173

ChatGPT citations carrying utm_source=openai in our two-engine research run of 4 September 2026, as published in who the AI actually cites and re-counted from our stored answers on 25 September 2026; Perplexity tagged none of its 391. From our own systems as read on 25 September 2026, that run collected ChatGPT's answers through OpenAI's API and Perplexity's through its own API. OpenAI's publishers FAQ, in the Internet Archive's capture of 13 September 2026, says ChatGPT adds utm_source=chatgpt.com to its referral URLs. The two routes differ. The run recorded cited URLs, not visits.

False information about your company in ChatGPT? What to do

You can't delete a ChatGPT answer, but you can report it. The steps ChatGPT, Google and Perplexity give, what each promises, and what to fix on your site.

$989

a month, the price Otterly's own list "The 25 Best AI SEO Tools in 2026", last updated 4 August 2026 (the Archive's capture of 5 June 2026, historical, shows it last updated 1 February 2026 with the same line), prints for an OtterlyAI plan it calls Pro. Otterly's pricing page, rendered on 25 September 2026, names no Pro plan and draws Lite at $29, Standard at $189 and Premium at $489 a month; the Internet Archive's capture of that page on 4 August 2026, the list's current last-updated date, carries the same three figures and no $989.

GEO case studies: the lift figures the vendors publish, and what each one measured

The 55 customer case-study pages linked from the indexes of AthenaHQ, Profound, Otterly, Goodie and Scrunch, read on 25 September 2026. 45 carry a headline lift figure; this post lists what each one counts, its starting value, window and engine, and who measured it, as each vendor's page states them.

7 of 45

headline lift figures on the customer case-study pages of AthenaHQ, Profound, Otterly, Goodie and Scrunch that print both a starting value and a time window for the change, as served and as rendered on 25 September 2026. The 19 pages that print a date of their own range from 15 April 2025, historical, to 3 September 2026; the other 36 print none.

What GEO agencies publish about their prices: seven agencies' own pages, read 25 September 2026

How much does GEO cost? Seven agencies' own pages, read 25 September 2026: six print a price for their own work (one of those six pages, RevvGrowth's, historical), as monthly tiers, 90-day sprints or one-off audits, and each prints a range for the market. Each figure with its page, its date where the page shows one, its scope and its term, and none averaged or ranked.

6 of 7

GEO agencies whose pages we read on 25 September 2026 that print a price for their own work. Two of the seven pages are historical: RevvGrowth's, dated 21 April 2026, and Fuel Online's, the one with no price of its own, read from an Internet Archive capture of 12 May 2026 because its live site answered with a bot check. The seven were named in our editorial plan, not sampled, so this is a count of these pages, not a share of agencies; and the six prices cover different work on different terms, so none of them is a market rate.

Training data or live search: why ChatGPT can know your brand in one mode and not the other

ChatGPT searches the web when it judges a question would benefit, by OpenAI's help page as archived on 19 September 2026, and otherwise answers without a search. What OpenAI, Anthropic, Perplexity and Google say about the two routes, what two free brand checks say they read, and which tools name the model behind an answer, in our read of 24 September 2026; other pages read 25 September 2026, OpenAI's and Perplexity's from Internet Archive captures.

6 of 19

AI visibility tools that name a model version on their own pages, in our read of 24 September 2026; we re-read the six vendors' pages on 25 September 2026 and each still carries its version string. The other thirteen named none on the pages read on 24 September. In our reading, an answer given without a search rests on one model's training, and Anthropic's models page, read 25 September 2026, prints training data cutoffs from July 2025 to June 2026 across its current models.

LLM brand monitoring vs social listening: what each one watches

Social listening collects public posts, articles and reviews that contain your brand's name; LLM brand monitoring asks AI engines buyer questions and records whether the answer names you and which pages it cites. Brand24's and Mention's own pages, read 25 September 2026, and our two-engine run of 4 September 2026.

45.0% vs 30.0%

of 240 brand-answer pairs had the brand's name in the answer text, and 30.0% had the brand's own site among the answer's sources: our research run of 4 September 2026 on two engines, ChatGPT and Perplexity, 24 brands in four B2B software categories, both engines pooled, re-counted 23 September 2026, as published in named but not cited. The second count is about an answer's sources. On the mention pages of Brand24 and Mention we read on 25 September 2026, the unit collected is a post, article or review that contains the name; Brand24 sells AI-answer tracking separately, through Chatbeat.

AI Mode vs AI Overviews, and where Gemini fits: Google's AI surfaces in Google's own words

What Google's own pages, read on 25 September 2026, say AI Overviews, AI Mode and the Gemini app are, with each page's date, and how two AI-visibility vendors list the Google surfaces on their pages as served that day: Semrush's own answer to which platforms it tracks names six, AI Mode and AI Overviews among them; Similarweb's names four for its AI Brand Visibility tracker, AI Mode among them and not AI Overviews.

6 vs 4

Engines named in each vendor's own answer to which AI platforms it tracks, as served on 25 September 2026: Semrush's FAQ answer to "What LLMs does Semrush track?" names six, among them Google AI Mode and Google AI Overviews; Similarweb's answer to "Which AI chatbots and search platforms does Similarweb track?" names four for its AI Brand Visibility tracker, among them Google AI Mode and not AI Overviews. Names on a page, not coverage we tested, and not a ranking.

How to choose the prompts you track: buying questions, not keywords

Which prompts to track for AI visibility: the buying questions in the 25 prompts we published on 4 September 2026, what SparkToro's study found about how people word the same need (28 January 2026, historical), and what fifteen AI-visibility vendors' pages say their prompt suggestions start from, read on 25 September 2026.

20 of 25

Buyer prompts in our research run of 4 September 2026, printed verbatim in our post on getting cited by ChatGPT and Perplexity, that name no vendor. The other five are head-to-heads, one per category, each naming three vendors, and that post says those three get a mention rate partly handed to them by the question. The twenty ask which product is best for a segment or with a qualifier, best at one capability, or right for a growing business; none is a bare keyword.

Does AI traffic convert better? The published conversion figures, with what each one counted

Eleven studies that print a conversion figure for AI-referred visitors beside a baseline, two figures that reach the web only through relays, and two pages with no conversion figure at all, read on 25 September 2026. Each with its unit, sample and window, because purchases, sign-ups, qualified enquiries and lead-to-close rates are four different things and the post ranks none of them.

14.2%

The mean conversion rate of AI-referred visitors to qualified enquiries across 312 IT and technology service firms from January 2025 to January 2026, against 2.8% for Google organic visitors to the same firms, in Opollo's benchmark report of 22 February 2026, historical. A page on superprompt.com dated 18 September 2025, historical, prints the same two figures for a different sample of 347 businesses and a different window, and neither page mentions the other on what we read. Every other figure in the table is on its own unit and does not convert into this one.

AI crawlers take more than they send back, on the latest figures: crawl-to-referral ratios, compiled

Crawl-to-referral ratios and AI-crawler volume figures from three Cloudflare posts, TollBit, technologychecker.io's readings of Cloudflare Radar's API, Fastly and Vercel, read 25 September 2026, each with its unit, its network, its window and the date its page prints; figures older than 90 days are marked historical, and nothing is compared across networks.

179 : 1

AI bot visits per human visitor referral from AI applications for European publishers on TollBit's network in the first half of 2026, from the executive summary its State of the Bots page displays, read 25 September 2026; the same summary prints 150 : 1 for the first quarter of 2026, historical by our 90-day rule, and 227 : 1 for the second, and says the numbers "reflect only identified bots".

Did search volume fall 25% by 2026? Gartner's prediction against the measured figures, each in its own unit

Gartner's press release of 19 February 2024 (historical) predicted that traditional search engine volume would drop 25% by 2026. Read on 24 September 2026: Google's own statements (queries at an all time high, 29 April 2026, historical; over 5 trillion searches a year, 21 May 2025, historical), Statcounter's referral share (Google 91.61% in February 2024, historical, and 91.02% in August 2026), SparkToro with Datos (searches per searcher up 21.64% in 2024, historical), SparkToro with Similarweb (68.01% of US Google searches without a click, January to April 2026, historical), Pew's click rates (July 2025, historical) and Alphabet's revenue and paid clicks (to June 2026), each in its own unit and none in Gartner's.

25%

the drop in traditional search engine volume that Gartner's press release of 19 February 2024 (historical) predicted for 2026, with no base year, no count and no method on the page. On the pages we read on 24 September 2026, no party publishes a 2026 query count; Alphabet's releases say queries were at an all time high in the quarter to March 2026 (29 April 2026, historical) and that AI features are driving Search query growth (22 July 2026).

How often do AI assistants get facts wrong? Seven accuracy studies, compiled with their units

Seven studies of how often AI assistants and AI search get facts wrong, read on 24 September 2026: the Tow Center, the BBC, the BBC-EBU study, NewsGuard, DeepTRACE, Liu et al. and Vectara's leaderboard, each with what it counted as an error, which engines and versions, the sample and the window. Six of the seven are historical, and none is averaged with another.

51% vs 37%

of BBC-reviewed news answers with at least one significant issue, December 2024 against May–June 2025: the BBC's own answers in both rounds, compared in the BBC-EBU study of October 2025, historical. 362 answers in the first round, 237 core and custom answers in the second. The report compares BBC-only data because its 22-broadcaster results are not directly comparable to the first round, and says this comparison, with some small differences in methodology, gives a sense of the direction of travel.

Which AI assistant do people use? Five published figures, five units, dated

Usage and share figures for ChatGPT, Gemini, Perplexity, Copilot and Claude from five sources, read on 24 September 2026: Statcounter's referral share, Similarweb's site ranking, Google's own monthly active users, the Reuters Institute's survey and Pew's. Each with its unit, sample and date, because no two of them count the same thing.

79.4%

ChatGPT, in one unit. 79.4% is its share of AI chatbot referrals to websites worldwide in August 2026, per the summary list on Statcounter's page read on 24 September 2026: clicks from a chatbot to a website, not people. The other four sources in the table use a second, third, fourth and fifth unit, and none of them converts into this one.

Do B2B buyers research software with AI? Four 2026 buyer surveys, every figure with sample and question

Four surveys of B2B buyers and AI published in 2026, read on 24 September 2026: Forrester (94% of nearly 18,000 business buyers; January 2026, historical), TrustRadius (63% of 1,862), G2 (71% of 1,076; surveyed March 2026, historical) and Gartner via Demand Gen Report (45%; 26 May 2026, historical), each with its sample, date and question where the page prints one, and the 72% ChatGPT figure, attributed to Forrester, which none of the six public Forrester pages we read carries.

94% vs 45%

two 2026 figures for buyers who used AI in a purchase, measured differently: Forrester's Buyers' Journey Survey of nearly 18,000 global business buyers (blog posts of 21 and 22 January 2026, historical) and Gartner's survey as printed by Demand Gen Report (26 May 2026, historical), whose sample the brief does not state. The gap is the question asked, not a contradiction: Forrester counts any AI in the buying process, Gartner counts GenAI in a recent purchase.

Do AI answers cite the pages that rank? Every published overlap figure, and what each counted

Overlap figures between AI citations and search rankings from fifteen study pages, read on 24 September 2026, each with its unit: 37.9% of Google AI Overview citations in the first ten result blocks (Ahrefs, March 2026, historical), about 17% in the organic top 10 (BrightEdge, February 2026, historical), 90% of AI Overviews citing at least one top-10 page (seoClarity, October 2025, historical), and 87% of ChatGPT search citations matching Bing's top results (Seer, February 2025, historical).

37.9% vs 90%

of AI Overview citations, against 90% of AI Overviews. Ahrefs found 37.9% of 4 million cited URLs, across 863,000 keyword results pages, in the first ten result blocks for the same query (2 March 2026, historical); seoClarity found 90% of 362,000 US desktop AI Overviews citing at least one top-10 page (October 2025, historical). Both read on 24 September 2026; the first counts citations, the second counts answers.

Is your CDN blocking AI crawlers? What Cloudflare's defaults do, per its own pages

Cloudflare's own pages, read 24 September 2026: which AI-bot settings exist, what each one blocks, and the default its Block AI Bots page, last updated 1 July 2026, says Cloudflare will set for domains added from 15 September 2026 — Training and Agent bots blocked on pages that display ads, Search allowed. Plus what a client with no JavaScript got from three B2B pricing pages the same day.

2 of 3 vs 2 of 30

of the three AI-bot behaviours Cloudflare's Block AI Bots page defines — Search, Agent, Training — will be blocked on pages that display ads under the default that page, last updated 1 July 2026 and read 24 September 2026, says Cloudflare will set for domains added from 15 September 2026; and 2 of the 30 B2B pricing pages we fetched the same day with no JavaScript answered with a bot challenge, one from Cloudflare and one from DataDome, neither of them to an AI crawler.

"Best X" lists as AI sources: what five studies found

Five studies of "best X" lists as AI sources, read on 24 September 2026, each with its sample, engines and date: 43.8% of ChatGPT source page types (Ahrefs, December 2025, historical), 21.9% of 1,056,727 citations (Wix and Peec, March 2026, historical), about 11% self-promotional (Peec, March 2026, historical), and a brand's own cited listicle leaving it out of the recommendation in 69% of citations (Lily Ray, June 2026, historical) and 43% of answers citing a conference page (Ahrefs, July 2026).

224 of 323

citations of a brand's own "best [category]" list in Google AI Overviews that left that brand out of the recommendation — 69% — across 80 B2B software queries read on 15 April, 15 May and 8 June 2026, as Lily Ray counted them in her post of 17 June 2026, historical, read here on 24 September 2026.

How to track your brand in ChatGPT yourself: a spreadsheet method, and where it breaks

How to track brand mentions in ChatGPT and Perplexity with a spreadsheet, taken from our run of 4 September 2026: 25 published prompts, one answer per prompt per engine, one row per brand per answer, two yes-or-no columns. Where it breaks: a single run's standard error is 0.370 and the minimum recommended is 7 runs per prompt per day (Schulte, Bleeker and Kaufmann, April 2026, historical); one citation decision in seven flipped on regeneration with the sources frozen (Selvam and Ghosh, September 2026).

1 run vs 7 runs

per prompt per engine. The first is what our method of 4 September 2026 used, 25 prompts on two engines; the second is what Schulte, Bleeker and Kaufmann (arXiv, 8 April 2026, historical) say a per-brand detection rate needs per day, because a single run's standard error is 0.370 and seven runs' is 0.081. The sheet below is the one-run method, with that limit stated at the step where it bites. The 0.370 is the paper's average over its own Swiss-German prompts, not a measurement of our run.

How much do AI answers change from one run to the next? Six studies, compiled with dates

Six studies of run-to-run variation in ChatGPT, Perplexity, Gemini and Google's AI answers, read on 24 September 2026: SparkToro and Gumshoe, three arXiv papers, Sielinski's statistical framework and Otterly's stability study, each with its own unit, sample size and date. Why ChatGPT gives different answers, in their numbers rather than ours.

7 runs vs 60 to 100

per prompt, in the two of six studies that recommend a run count. Schulte, Bleeker and Kaufmann (arXiv, 8 April 2026, historical) put the standard error of a single run's per-brand detection rate at 0.370 and reach 0.081 at seven runs; Rand Fishkin (SparkToro, with Gumshoe's Patrick O'Donnell, 28 January 2026, historical) writes "usually at least 60-100X" to know an AI's set of recommendations. Different units — one brand's rate against the whole list — which is why the two figures do not contradict each other.

AI crawlers in one table: what OpenAI, Anthropic, Perplexity, Google, Apple, Common Crawl and Meta document

Seven vendors' own crawler pages, read 24 September 2026 and put in one table: 29 user-agent names, what each vendor says each is for, and what it says about robots.txt, quoted. Five names carry a stated exception to robots.txt; one of the seven pages mentions JavaScript.

3 of 4 vs 1 of 7

of the four agents these seven vendor pages describe as fetching a page at a user's request — ChatGPT-User, Perplexity-User, Claude-User and Meta-ExternalFetcher — three say on their own vendor's page that robots.txt may not apply to them, as read 24 September 2026; and one of the seven pages, Apple's, says anything about JavaScript or rendering.

Which AI-visibility studies could you check? Sample, prompts, raw data

All 61 study-shaped posts on the Profound, Otterly, Peec and AthenaHQ blogs, opened and rendered on 24 September 2026 and asked four questions: 31 state a sample size, 21 name a month, none of the 33 that report a measurement prints its full prompt set, none links raw data. Two of our own posts, asked the same, score yes, yes, yes and no.

0 of 61

study-shaped posts on the Profound, Otterly, Peec and AthenaHQ blogs that link the raw data behind their figures — a download, a public sheet or a repository — as served and as rendered on 24 September 2026. Two of our own posts on the run of 4 September 2026 score the same on that question.

Do the sites ChatGPT and Perplexity cite block their crawlers?

On 24 September 2026 we read the robots.txt of the 269 domains that ChatGPT and Perplexity had cited in our run of 20 prompts on 4 September 2026: 253 were readable and 7 disallow an OpenAI or Perplexity bot on every path. Three sites ChatGPT cited disallow GPTBot on every path and none disallows OAI-SearchBot on every path; 19 of Perplexity's 391 citations went to reddit.com, whose robots.txt disallows every crawler.

3 of 81 vs 0 of 81

domains ChatGPT cited on 4 September 2026 whose robots.txt, read 24 September 2026, disallows GPTBot on every path — against 0 of the same 81 disallowing OAI-SearchBot, the crawler OpenAI's page, read 24 September 2026, says a site must not opt out of to be shown in ChatGPT search answers. 81 is the number of ChatGPT's 88 cited domains whose robots.txt we could read that day.

What is AI visibility, and how is it measured?

AI visibility, on five vendors' own pages read 24 September 2026, is a share of answers or prompts naming the brand at Profound, Peec and Otterly, a 0–100 score at Semrush and a set of counts at Ahrefs; all five also count the brand's pages among the sources. Our run of 4 September 2026, two engines and 240 brand-answer pairs: 45.0% named, 30.0% cited.

45.0% vs 30.0%

of 240 brand-answer pairs had the brand's name in the answer text, and 30.0% had the brand's own site among the answer's sources — ChatGPT and Perplexity, 24 brands in four B2B software categories, run 4 September 2026, re-counted 23 September 2026, as published in named but not cited. Every vendor in the table keeps a count of each kind, and neither of ours is any vendor's figure.

Eight questions to ask an AI visibility tool before you buy, and what nineteen tools' own pages say to each

Eight questions to put to an AI visibility tool's own pages before you buy, each with a count from our reads of nineteen tools on 23 and 24 September 2026 and a link to the post that carries it: what it shows you, how it collects answers, which model, how often, what a prompt with no AI Overview, or a failed fetch, counts as, whether its pages agree on the engine count, and what it costs. No count here is new.

0 of 19

AI visibility tools that state, on the pages read on 24 September 2026, how a fetch that fails enters their figures — dropped, retried, counted as an answer with nobody in it, or left as a gap — per our post on what nineteen of them say about how they get their answers. Conductor's docs name the case as a status and Athena's API reference defines its denominators as successfully analyzed responses; neither is a rule for the count. One of nineteen states a rule for a prompt with no AI Overview: Otterly, in KPI definitions dated 4 September 2026.

Can you see the answer, and who got named instead? What nineteen AI visibility tools' pages say

Nineteen AI visibility tools, read from their own docs, help centres and product pages on 24 September 2026: sixteen say you can open the answer an engine gave, eleven say they list brands you did not add, and eleven say they show the URLs each answer cited.

11 of 19

AI visibility tools whose own pages, read on 24 September 2026, say the product lists brands or competitors you did not declare — the name that came out instead of yours. Five say they show which brands were mentioned, or a competitor set found automatically, without saying whether the names come from the answers; three say nothing on the pages read; none says the opposite on the pages read. Sixteen of nineteen say you can open the answer text; eleven say they show the URLs each answer cited.

The vendors' own "best AI visibility tools" lists: 14 of 14 put the author first

Every best-of and top-N list of AI-visibility tools on the Profound, Otterly, AthenaHQ, Peec and Geoptie blogs, opened and rendered on 24 September 2026: where the author's own tool sits, how many tools each names, and whether the 179 prices they print for other tracked tools are on those tools' own pages.

14 of 14

best-of and top-N lists of AI-visibility tools on the Profound, Otterly, AthenaHQ, Peec and Geoptie blogs that name two or more tools, and put the author's own tool first — every post whose slug or title carries best or top, a plural tool noun and an AI-visibility term, rendered in a browser on 24 September 2026.

30 B2B software pricing pages: how many show a price before JavaScript runs?

Thirty B2B software companies' pricing pages, fetched with curl and then rendered in headless Chrome on 24 September 2026: thirteen serve a plan price before any script runs, none draws one only with JavaScript, nine show none either way, five have no pricing page, and three answered a script with a 403. What OpenAI's and Perplexity's crawler pages say about JavaScript: nothing.

13 of 30

B2B software companies whose pricing page carries a plan price in the HTML it serves, before any script runs, measured 24 September 2026 with curl and a browser user agent and then rendered in headless Chrome with JavaScript on. Not one of the thirty draws its price only with JavaScript; nine have a pricing page with no figure, five have no pricing page, three answered a script with a 403.

HeardOf vs Profound: what each one measures, and what Profound does better

A Profound alternative. Profound's pricing page, drawn in a browser on 24 September 2026, shows no plan price for a brand — Trial Free, Enterprise Custom — and, after a click on its agencies tab, $399 a month for an added client workspace. Its docs leave answers that name no brand out of its visibility score; ours count them, per our post on which engines we check, as of 23 September 2026. What Profound does better, and the three things our post prints that its pages do not.

$399/mo

for one added client workspace on Profound's Agency Growth plan, one of two add-on figures its pricing page draws — the other is $199 a month for five trial workspaces — in a browser on 24 September 2026, and only after clicking For agencies; as the page opens it shows Trial, Free, and Enterprise, Custom, and the HTML it serves carries no figure on either tab. Profound's figure, not ours; what HeardOf charges is on our own site and not in this post.

HeardOf vs Peec AI: what each one measures, and what Peec does better

A Peec AI alternative, compared honestly. Peec's pricing page, drawn in a browser on 24 September 2026, shows Starter at $95 a month, Pro $245 and Advanced $495 on monthly billing, in US dollars as drawn for us; the HTML it serves carries none of those figures. SE Ranking's review lists $80 on annual billing (modified 8 September 2026). Each self-serve plan runs 50 to 350 prompts you keep, daily, on three of six engines. HeardOf, per our post on which engines we check as of 23 September 2026, asks four engines 40 buyer prompts each. What Peec does better, and the two counting rules our post prints that the Peec pages we read do not.

$95/mo

for Peec AI's Starter plan on monthly billing, as Peec's pricing page draws it in a browser on 24 September 2026, in US dollars as drawn for us; the HTML the page serves carries none of its figures, and a script draws them in. The same $95 is on Peec's ai-instructions page, whose header says pricing was confirmed live on 10 July 2026, and in SE Ranking's review of Peec, modified 8 September 2026, which lists $80 a month on annual billing. Peec's figures and a review site's, not ours; what HeardOf charges is on our own site and not in this post.

AI visibility tools: what nineteen of them say about how they get their answers

Nineteen AI visibility tools, read from their own docs, help centres, knowledge bases and product pages on 24 September 2026: thirteen say how they collect answers, six name a model version, one states how a prompt counts when Google shows no AI Overview, none says how a failed fetch is counted, and six give two different engine counts on their own pages.

13 of 19

AI visibility tools whose own pages say how they collect answers — through the chat interface, through an API, or both — when read on 24 September 2026. Six say nothing on the pages we read. Six of the nineteen name a model version; one states how a prompt counts when Google shows no AI Overview; none states how a failed fetch enters its figures.

HeardOf vs Otterly: what each one measures, and what Otterly does better

An Otterly alternative, compared honestly. Otterly's Lite plan is $29 a month, or $25 on annual billing, for 15 prompts on four AI engines, with three more engines as paid add-ons, on its own pricing page read 23 September 2026. HeardOf, per our post on which engines we check as of the same date, asks four engines 40 buyer prompts each and scores you and up to three competitors you name. What Otterly does better — engines, countries, daily tracking, sentiment and position — and what its help center prints that ours does not.

$29/mo

for Otterly's Lite plan — 15 prompts, four AI engines, monthly billing — on its own pricing page, read 23 September 2026; $25 a month on annual billing, and $189 and $489 a month for Standard and Premium on the same page. Otterly's figures, not ours; what HeardOf charges is on our own site and not in this post.

Semrush AI Visibility Toolkit vs HeardOf: what each one measures, and what Semrush does better

Semrush's AI Visibility Toolkit is $99 a month per domain, billed annually, on its own pricing page read 23 September 2026, and its AI-visibility solutions page names six AI platforms. HeardOf, per our post on which engines we check as of the same date, asks four engines 40 buyer prompts each and scores up to four brands you name. What Semrush does better — platforms, SEO data, the brand — and the one comparison the evidence did not support.

$99/mo

per domain, billed annually, for Semrush's standalone AI Visibility Toolkit on its own pricing page, read 23 September 2026. The cheapest Semrush plan that bundles it with the SEO tools is $165.17 a month billed annually, or $199 monthly, on the same day. Semrush's figures, not ours; what HeardOf charges is on our own site and not in this post.

574 posts: what the AI-visibility category writes about itself

A census of five AI-visibility vendor blogs — Profound, Otterly, AthenaHQ, Peec and Geoptie — read from their own sitemaps and pages on 23 September 2026, with every method printed: 574 posts, how many were revised, how often each names a competitor, 50 templated industry pages, and what the category writes about. Every count can be re-run.

14 of 155

posts published before 1 June 2026 and modified on or after it, on the two AI visibility vendor blogs — Otterly and Geoptie — whose modification dates can be read post by post; counted 23 September 2026 from Otterly's WordPress API and Geoptie's structured data. The other three sites carry no per-post modification date we could use, and the post says why for each.

Named but not cited: being mentioned by an AI is a different job from being its source

On 4 September 2026 we scored 240 brand-answer pairs from ChatGPT and Perplexity in four B2B software categories two ways: was the brand named in the answer, and was its own site among the sources. Named: 45.0%. Cited: 30.0%. Thirty-seven pairs named a brand and linked to somebody else; one did the reverse.

45.0% vs 30.0%

of 240 brand-answer pairs had the brand's name in the answer text; 30.0% had the brand's own site among the answer's sources — ChatGPT and Perplexity, 24 brands in four B2B software categories, 4 September 2026, re-derived from the stored answers on 23 September 2026.

What nineteen AI-visibility tools charge, and the ten things we could not verify

AI-visibility tools pricing, read from all nineteen vendors' own pages on 23 September 2026 and rendered in a browser on 24 September: fifteen show a plan price served or drawn, four show no plan price either way, and no third-party number is passed off as a vendor's. Two prices changed since our first read, one may have, and five rows were our own earlier read being incomplete. The ten things we could not verify on the first read are listed by name; two were settled by the second.

15 of 19

AI-visibility tools whose own pricing page showed a figure when we read it on 23 September 2026 or drew one when we rendered it on 24 September; Peec's is the one only a browser shows. The four that show none are Profound, Bluefish, Conductor and BrightEdge; the third-party estimates that circulate for them are attributed in the table and are not printed as theirs.

Does llms.txt actually work? What four vendor posts claim, and what their own sites serve

Four posts from three AI-visibility vendors — Profound, Peec and Otterly — recommend, doubt or measure llms.txt. On 23 September 2026 we curled the three sites: one serves the file at its root, all three on their docs host, and Profound's root URL is a 404 that its own sitemap and robots.txt point at. Every command is in the post.

1 of 3

vendors that published a post about llms.txt serve the file at their root domain — Otterly. All three serve one on their developer-docs host, and Profound's root URL is a 404 that its own robots.txt and sitemap point at. Measured with curl -L on 23 September 2026.

Muck Rack's 84% is not 84% journalism. Here is what that number counts

Muck Rack's What Is AI Reading?, May 2026, reports earned media at 84% of AI citations. Journalism alone is 27%, owned media 13.7%, and the May 2026 edition measures ChatGPT, Claude and Gemini but not Perplexity. We read seven Muck Rack pages on 21 September 2026 and set the figure against our own run of 4 September 2026, where ChatGPT put 44.5% of its citations on vendors' own sites.

84% vs 27%

of AI citations are "earned media" in Muck Rack's May 2026 report, but journalism alone is 27%. The rest of the category is encyclopedias, academic and government sources, and third-party corporate content.

Who the AI actually cites — and the two engines do not agree

On 4 September 2026 we asked ChatGPT and Perplexity the same 20 buyer questions in four B2B software categories and kept every source they cited: 564 citations, 269 domains. 44.5% of ChatGPT's pointed at a vendor's own site; 9.2% of Perplexity's, in the same run. Reddit was the top domain that day at 3.4%, and ChatGPT cited it zero times.

44.5% vs 9.2%

of ChatGPT's citations pointed at a measured vendor's own site — one of the 24 we measured — against 9.2% of Perplexity's — 564 citations from 40 answers in four B2B software categories, 4 September 2026, re-derived from the stored answers on 19 September 2026.

The four engines we ask, and what changed when Google AI Overviews became one of them

HeardOf checks four engines — ChatGPT, Perplexity, Gemini and, since 23 September 2026, Google AI Overviews — 40 buyer prompts each, 160 asked per audit. What the fourth is read from, what a prompt with no overview counts as, why earlier scores were not recomputed, and what 19 competing tools say they check.

160

answers per audit when all four engines answer — 40 buyer prompts, each put to ChatGPT, Perplexity, Gemini and Google AI Overviews. The fourth engine was switched on 23 September 2026; audits before it read up to 120 answers on three engines, and those scores stand as they were.

Want your own numbers instead of ours?

Up to forty of your buyers’ questions, four engines, who got named. About 15 minutes.

Start Monitor