AI SEO Tools Benchmark: method
How we test: every question, rule and limit.
Everything you need to repeat the study or argue with it: the questions, where and how we asked them, the rules that count a mention, the scoring weights, the limits of a one-day sample, and every change we made along the way.
We re-run the study monthly; each edition keeps its own record.
The method record
- Edition
- Edition 1, October 2026
- Sample
- 1,199 answers, collected on 16 October 2026 (India time); one of them a ChatGPT test answer from 13 October
- Asked from
- United States, in English
- Brand list
- Dictionary v3, frozen on 16 October 2026
- Page-type rules
- v3
- Weights set
- 16 October 2026
- Changes recorded
- 18 dated entries, all below
- Written by
- Chandan Bakshi, Founder, SEO Informatica Labs
The questions, in full.
The same 25 questions, word for word, on every surface: the questions a buyer asks before choosing an AI SEO tool. Like buyers, they cover both meanings of AI SEO, AI that does SEO work and visibility in AI answers; our guide explains what AI SEO is.
- What are the best AI SEO tools right now?
- Which AI SEO tools are worth paying for?
- What AI SEO software do agencies use?
- What are the best free AI SEO tools?
- Which AI tools should a small business use for SEO?
- What is the best tool to track my brand's visibility in ChatGPT?
- Which tools show whether my website is cited in AI answers?
- What are the best AI visibility tools for ChatGPT, Gemini and Perplexity?
- How can I monitor brand mentions across AI search engines?
- Which tools track citations in Google AI Overviews?
- What are the best generative engine optimization tools?
- What are the best answer engine optimization tools?
- Which tools help my brand get recommended by ChatGPT?
- What tools help optimize content for AI search?
- Which GEO platforms do enterprise marketing teams use?
- Which AI visibility tool do SEO experts recommend most?
- What is the cheapest AI visibility tracking tool that is actually good?
- Which AI visibility tool is best for a small agency with ten clients?
- Which AI visibility platform is best for a large brand?
- Which AI visibility tool has the most accurate data?
- What is the best AI tool for SEO keyword research?
- What is the best AI tool for writing SEO content?
- Which AI tools help with technical SEO audits?
- What is the best AI tool for local SEO?
- Which tool shows me which websites ChatGPT recommends in my industry?
Where and how we asked.
We collected every answer through DataForSEO, a data provider that asks these products for us from a set location and language. ChatGPT and Perplexity give a different answer each time, so we asked each question many times there; Google shows one AI Overview and one AI Mode answer per search, so we asked each question once.
| Surface | How we asked | Model recorded | Runs per question | Answers | Collected (UTC) |
|---|---|---|---|---|---|
| ChatGPT | the ChatGPT app, read as a user sees it, with web search on. DataForSEO AI Optimization API, ChatGPT LLM Scraper (task_post, standard queue). | gpt-6 | 30 | 750 answers | From 15 October 2026, 19:48 UTC to 16 October 2026, 07:47 UTC, plus one test answer on 13 October 2026, 13:25 UTC |
| Perplexity | Perplexity's Sonar model through its API, which searches the web by default, from the United States. DataForSEO AI Optimization API, Perplexity LLM Responses (live). | sonar | 16 | 400 answers | From 16 October 2026, 07:48 UTC to 16 October 2026, 07:51 UTC |
| Google AI Overviews | the Google search results page, desktop, with the AI Overview loaded. DataForSEO SERP API, Google Organic Live Advanced, with load_async_ai_overview on. | Not named by the surface | 1 | 24 overviews from 25 searches | From 15 October 2026, 19:50 UTC to 15 October 2026, 19:55 UTC |
| Google AI Mode | Google AI Mode, desktop. DataForSEO SERP API, Google AI Mode Live Advanced. | Not named by the surface | 1 | 25 answers | From 16 October 2026, 07:49 UTC to 16 October 2026, 07:52 UTC |
Claude and Gemini were tested and left out. Claude (claude-sonnet-5-5): one answer cost about 44 times a ChatGPT answer, so 1 answer per question would fit its allowance against the 10 we require; Gemini (gemini-3.6-flash): one answer cost about 32 times a ChatGPT answer, so 2 answers per question would fit its allowance against the 10 we require.
A one-day sample.
Every counted answer but one was collected on 16 October 2026 (India time), from 15 October 2026, 19:48 UTC to 16 October 2026, 07:52 UTC, from the United States in English. The exception is the first ChatGPT answer, a test answer collected on 13 October 2026, 13:25 UTC, which is counted with the others.
- One day of answers is a sample. Answers change from day to day, so a rate here describes that day and is not a forecast.
- Google AI Overviews and AI Mode were asked once per question, so their rates rest on a few dozen answers and a single answer moves them a lot.
- We asked from the United States in English. Answers in other places and languages may differ.
- A mention is a name in an answer. It does not say whether the answer recommended the tool or warned against it.
How a mention is counted.
The brand list is dictionary v3, frozen for this edition: no brand is added or removed until the next one.
The rules, in plain words:
- Mention. A brand is mentioned when its name, or another name we recorded for it, appears in the answer's text as a whole word, in any capitals. For a name of two or more words the space is optional, so Surfer SEO also matches SurferSEO.
- Citation. Counted separately from mentions: one of the brand's websites, or a page under it, is among the sources the answer links to. A brand can be cited without being named, and named without being cited.
- Names that are also ordinary words. For brands such as Profound, Conductor or Gauge, the bare word counts only when it is capitalised mid-sentence, sits next to a word such as tool, platform or software, starts a list item followed by a colon or a dash, or the brand's website is among the sources. Every other hit goes to a person, who answers yes or no.
- Exact capitals. CITED, which is also an everyday word, counts only in its exact capitals.
- Status. Every brand carries a status: live, renamed, acquired, acquired but still sold under its own name, acquired and no longer sold, parked or unclear. Matching uses every brand; our published lists show only the brands still sold.
Dictionary versions.
- v1: 13 October 2026: the starting list we set for the study (89 brands), plus 38 websites filled in during setup.
- v2: 16 October 2026: SE Ranking merged with SE Visible; 21 missing websites looked up; moved domains added for Scrunch, Promptwatch, ZipTie and Bluefish; the spellings OtterlyAI and Rank Prompt; a status for every brand; 18 products added that were named in 3 or more answers; new rule: an ordinary-word name at the start of a list item followed by a colon counts. Later on 16 October 2026: Scrunch set to "acquired, live" (Sitecore).
- v3: 16 October 2026: 27 products added that were named in 5 or more answers across all engines, each with website, category and status, marked added v3; HubSpot gains the other name AI Search Grader; CITED is matched only in capitals. Frozen for edition 1. Later on 16 October 2026: Semrush set to "acquired, live" (Adobe, completed 28 April 2026); owners recorded for Semrush and Scrunch. XFunnel set to "acquired, dormant" (HubSpot, agreement announced 31 October 2025), so 64 AI visibility tools are live. No brands added or removed.
The page types, rules v3. A cited page gets the first type that fits, checked in this order: video, review site, community, a tool maker's own site, third-party list, news, other. The types are:
- Vendor product or pricing page
- Vendor-written list
- Third-party "best tools" list or comparison
- Review site or software directory
- Community (Reddit, forums, social)
- News or trade press
- Video
- Other
v1, seven page types; v2, vendor pages split into product or pricing pages and vendor-written lists; v3, "tools" makes a vendor page a list only alongside another list word (all 16 October 2026).
How the tools are scored.
Each live AI visibility tool is scored out of 100. In this edition only the openness score is filled in.
| Part | Points | What it covers |
|---|---|---|
| Coverage | 15 | Which AI engines it tracks, and on which plans |
| Transparency | 15 | Disclosed run counts and sampling, and whether the models and the surface (app or API) are named |
| Price | 15 | A published starting price with plan limits, and a free tier |
| Freshness | 10 | How often the data updates |
| Exports | 5 | Data exports and a documented API |
| Openness score | 60 | The openness score rates what a vendor publishes about its product (engines, method, price, data freshness, exports), not how good the product is. |
| Accuracy against the measured mention rates | 30 | Reads "pending access" for every tool in this edition. |
| Setup time | 10 | Reads "pending access" for every tool in this edition. |
Accuracy and setup time need a tool account set up with our questions, so we can set the rates the tool reports against the rates we measured, and time the setup. Until then they read pending access. A pending score is not a low score, and no tool is ranked on accuracy while it is pending.
An open invitation to every maker: give us test access and we will score accuracy and setup time on the same questions, stating the date access was granted, the plan it ran on and the dates of the answers it was compared with.
The level each recorded fact can reach
- Starting price and plan limits: 0, contact sales or no price; 1.0, public price and concrete plan limits; 0.6, public price, limits vague; 0.3, a 'from' price only, or a price only after a form or demo.
- Engines tracked: 0, not stated; 1.0, named engines, stated per plan where plans differ; 0.6, named engines, plan differences unclear; 0.3, vague, for example 'all major AI engines'.
- Disclosed run counts and sampling: 0, nothing disclosed; 1.0, runs per prompt per period and how results are aggregated; 0.5, only a cadence or only a run count.
- Model and surface disclosed: 0, neither; 1.0, names the models and says app or UI capture versus API; 0.5, one of the two.
- Data freshness: 0, not stated; 1.0, stated cadence daily or better, or on demand; 0.6, stated weekly or slower.
- Exports and API: 0, not stated; 1.0, export and a documented public API; 0.6, one of the two; 0.3, only as enterprise or custom, no docs.
- Free tier: 0, none; 1.0, a free plan with no time limit, or a free checker that returns an AI visibility result (not only a site-readiness grade) with no login and no email; 0.6, free trial with no card; 0.2, trial needing a card, or demo only.
Facts come only from vendor-owned pages that were read (site, subdomains, docs, help centre, changelog, official blog, GitHub). Where two vendor pages contradict each other, the lower level is used and both pages are named. A level above 0 needs a source. What the browser runs showed on 16 October 2026 corrects free tier where it differs from the vendor's wording.
The self-naming and review rules.
A free checker's prompt that names the brand it scores cannot measure that brand's visibility, so we take it out: 29 of the 667 prompts the checkers showed were self-naming.
Some brand names are also ordinary words. A hit on one of those counts only in a clear product context; every other hit goes to a person, who answers yes or no. A yes for a brand in a list-like layout also counts that brand's other uncertain hits in the same layout. Every hand answer is kept with the data.
Which tool pages search engines may list.
A tool page is indexable when the answers named the tool at least 10 times across the four surfaces (of 1,199 answers), or its free checker returned a result, or it has a recorded ownership change.
Indexable means search engines may list the page in their results. A tool page that meets none of the three carries a noindex tag, which asks search engines not to list it; the page is still published and linked from the table of every tool.
In this edition 81 of the 124 tool pages are indexable and 43 carry a noindex tag.
Scope notes.
Printed in full, as recorded with the data.
The scope decisions we wrote down for the AI SEO Tools Benchmark, kept and versioned with the data.
Scoring the AI visibility tools
Each live AI visibility tool is scored out of 100: openness score 60, accuracy 30, setup time 10.
The openness score rates what a vendor publishes about its product (engines, method, price, data freshness, exports), not how good the product is.
- The openness score (60) is scored now, for every live AI visibility tool, from the vendor's own pages only: coverage 15, transparency 15, price 15, freshness 10, exports 5. Every fact carries the page it came from and the date it was read; the scorecards are in the downloads on the data page. The weights, and which recorded criteria each part covers, are kept with the data.
- Accuracy (30) needs a tool account set up with our questions: the same 25 buyer questions, asked from the United States in English, so that the mention rates the tool reports can be set against the rates this study measured. A public page cannot show this, and a free checker cannot either, because it asks its own questions, not ours.
- Setup time (10) is measured in that same account as active setup time: the timer starts when setup begins after access is confirmed and stops when the 25 questions are saved and the first collection is requested or scheduled. Waiting time is recorded separately and not scored.
- Until a tool's account is set up with our questions, its accuracy and setup time read "pending access". A pending score is not a low score, and no tool is ranked on accuracy while it is pending.
- We invite any maker to grant test access. Tools are scored on accuracy as access arrives. Each accuracy score carries the date access was granted, the plan it ran on, and the dates of the answers we measured that it is compared with.
Free public checkers
- A free checker that returns a result without an account or email is run on the same five brands (Semrush, Otterly, Ahrefs, Profound, Peec AI) plus SEO Informatica as the control, with identical inputs on every checker.
- Its verdict is compared with our measured mention rates. This shows whether the checker's headline agrees with what buyers were actually shown on 16 October 2026. It is not an accuracy score for the paid product, and it does not count toward the 30 accuracy points.
- A checker that stops at a "verify you are human" check (a CAPTCHA, Cloudflare Turnstile or reCAPTCHA) is recorded as "needs a human click" and left for a person to run by hand. It is never worked around. On 16 October 2026 that was 8 checkers: Semrush, Ahrefs, Indexly, OpenLens, VisibilityTool.ai, Searcherries, SEOJuice and Parse.gl.
- When a checker does not put the control low, the prompts it showed for the control decide what that means, in its control verdict: "mistake" if they were about AI SEO or AI visibility tools, the category we measured; "different category" if they were about something else; "unverifiable" if it showed no prompts. Agency questions ("best AI SEO agency") count as a different category, because our 25 questions ask about tools.
- A self-naming prompt contains the brand being scored, in any spelling (for example "What is Seoinformatica?"). It cannot measure visibility, so a checker's result on it reads "not a visibility measurement" and it is left out of the control verdict. Where a checker shows every prompt and a result for each, the self-naming ones are removed and its score is recomputed on the rest; where it gives one score over several prompts, the row is flagged instead. On 16 October 2026, 29 of the 667 prompts the checkers showed were self-naming.
Every change, in full.
One dated entry for every change that altered a published number or rule, oldest first, with what moved.
-
16 October 2026, 02:33 India time: Brand dictionary v2
- What changed: SE Ranking and SE Visible became one brand (SE Ranking, with SE Visible and "SE Ranking AI visibility" as other names). The 21 brands with no website got one, looked up and marked "added". New domains were added beside the old ones for Scrunch, Promptwatch and ZipTie, whose old domains now redirect, and the right domain was added for Bluefish (bluefishai.com; bluefish.ai does not answer). The spellings "OtterlyAI" and "Rank Prompt" were added. Every brand got a status (live, renamed, acquired, parked or ambiguous), and published lists use live brands only. 18 products named in 3 or more answers were added, marked "added v2". The dictionary went from 89 to 107 brands.
- Why: a brand with no website, or with only its old domain, cannot have its citations counted; a missed spelling undercounts mentions; and a published list should not rank products that were renamed, parked or bought.
- Numbers that moved:
- Fixed before we wrote up the first batch of answers (01:35 on 16 October): Scrunch's ChatGPT citations went from 23 to 56 of the 250 answers once scrunch.com was added (and its mentions from 58 to 59), and Otterly's ChatGPT mention rate went from 52% to 63% once "OtterlyAI" was matched. The same spelling fix moved Otterly in AI Overviews from 6 to 11 of the 24 overviews.
- Statuses: 101 live, 1 renamed, 1 parked, 2 acquired, 2 ambiguous.
- The top 5 by ChatGPT mention rate did not change (Semrush 78%, Otterly 63%, Ahrefs 58%, Profound 58%, Peec AI 52%). Google Search Console, one of the 18 added, came sixth by ChatGPT mention rate (29%).
-
16 October 2026, 02:33 India time: Ordinary-word names count at the start of a list item followed by a colon
- What changed: a brand whose name is also an ordinary word (Profound, Conductor, Gauge and the like) now counts when it starts a list item and is followed by a colon, for example "- Profound: tracks AI answers". The 9 Profound hits waiting for a person's answer were also answered yes by hand.
- Why: those 9 hits were clearly the product, written as a list item, but the stricter check for ordinary words held them back.
- Numbers that moved: Profound in Google AI Overviews went from 3 to 12 of the 24 overviews. Of the 9 hits, the new rule counts 6 by itself and the other 3 count through the hand answers.
-
16 October 2026, 02:53 India time: The list-item rule widened to a colon or a dash
- What changed: the ordinary-word rule above now also accepts a dash after the name, and a numbered line written in bold counts as a list item. A plain hyphen needs a space on each side, so "Profound-style" still does not count. The waiting hit in question 16 was also answered yes.
- Why: ChatGPT's answer to question 16 wrote a bold numbered line, "1. Profound", then a dash, then "Best for enterprise SEO": clearly the product, with a dash instead of a colon.
- Numbers that moved: only one count. Profound in ChatGPT went from 144 to 145 of the 250 answers, all in question 16 (9 to 10 of 10). Every other count, including every AI Overview count, stayed the same. Hits waiting for a person's answer went from 5 to 2.
-
16 October 2026, 02:53 India time: Page-type rules v2, vendor pages split in two
- What changed: the page type "vendor's own site" was split into "vendor product or pricing page" and "vendor-written list". A vendor page counts as a list when its title or address contains blog, best, top, tools, compare, alternatives or a year (a blog. subdomain counts as part of the address).
- Why: until then, a list written on a vendor's own blog counted as the vendor's site, so the vendor share mixed product pages with vendor-written rankings.
- Numbers that moved:
Page type ChatGPT before ChatGPT after AI Overviews before AI Overviews after Vendor's own site 64.4% split 10.1% split Vendor product or pricing page 57.1% 1.8% Vendor-written list 7.3% 8.3%
Shares of 2,960 ChatGPT links and 168 AI Overview links. The other page types did not change. Two new top-10 lists of the most-cited vendor-written list pages, one for ChatGPT and one for AI Overviews, were added to our written summary of the results.
-
16 October 2026, 12:39 India time: Page-type rules v3, the word "tools"
- What changed: "tools" now makes a vendor page a list only alongside another list word (blog, best, top, compare, alternatives or a year). On its own it no longer counts.
- Why: 75 of the 229 vendor-written list links (74 of 215 in ChatGPT, 1 of 14 in AI Overviews) qualified on "tools" alone, and nearly all of them were product, free-tool or help pages. 4 of the 10 most-cited "vendor-written lists" were not lists at all (Google Keyword Planner, Ahrefs Free, Bing Webmaster Tools and an Ahrefs help page).
- Numbers that moved:
Page type ChatGPT before ChatGPT after AI Overviews before AI Overviews after Vendor product or pricing page 57.1% 59.6% 1.8% 2.4% Vendor-written list 7.3% 4.8% 8.3% 7.7%
All four (Keyword Planner, Ahrefs Free, Bing Webmaster Tools and the Ahrefs help page) left ChatGPT's top-10 vendor-written lists, which are now all genuine lists. Brand counts did not change.
-
16 October 2026, 13:28 India time: New status "acquired, live", first used for Scrunch
- What changed: a brand that was bought but still runs as its own product gets the status "acquired, live" and stays in every ranking. Scrunch was set to it, with a note naming Sitecore as the buyer (the owner got a field of its own later, at 14:28 the same day).
- Why: Scrunch's own blog post says it is now a Sitecore company and keeps operating, so it is still a product a buyer can choose.
- Numbers that moved: none. Re-running the count with Scrunch set back to "live" gives the same numbers; only its status and the tally of statuses change.
-
16 October 2026, 13:38 India time: Brand dictionary v3, frozen for edition 1
- What changed: 27 products named in 5 or more answers across all engines were added, marked "added v3", each with website, category and status. HubSpot gained "AI Search Grader" as another name. CITED is matched only in capitals. The dictionary went from 107 to 134 brands and was frozen: no brands are added or removed for edition 1.
- Why: v2 was built from the first 250 ChatGPT answers and the 24 AI Overviews. Once 500 more ChatGPT answers, 400 Perplexity answers and 25 AI Mode answers were in, products that v2 had missed showed up often enough to count: most were found in Perplexity, and about a third (for example Adobe Brand Visibility, CITED, CitedSpy, Searcherries) only in the newer ChatGPT answers. 41 answers call HubSpot's free tool "HubSpot AI Search Grader". "cited" is an everyday word in these answers, so CITED needs its exact capitals.
- Numbers that moved: HubSpot in Perplexity went from 109 to 125 of 400 answers, all from the new name (re-running the count without it gives 109). The 27 new products got their own counts, mostly in Perplexity. No other existing brand count changed through the dictionary.
- Page types also moved, because the new brands' websites now count as vendor sites:
Page type ChatGPT Perplexity AI Overviews AI Mode Vendor product or pricing page 59.6% to 60.7% 1.6% to 4.0% unchanged unchanged Vendor-written list 4.7% to 5.4% 39.6% to 42.9% 7.7% to 9.5% 5.9% to 6.9% Third-party "best tools" list 26.2% to 25.5% 47.9% to 46.0% 48.8% to 48.2% unchanged Other 8.3% to 7.2% 8.8% to 5.2% 14.3% to 13.1% 6.9% to 5.9% - Cituna, Bloomiro and Beamtrace posts entered ChatGPT's top-10 vendor-written lists.
- Page types also moved, because the new brands' websites now count as vendor sites:
-
16 October 2026, 13:38 India time: A person's yes teaches a pattern
- What changed: a person's yes for a brand in one layout now also counts that brand's other uncertain hits in the same layout. The only pattern then was Profound at the start of a list item, which counted 5 more rows by itself. Each pattern learned is kept with the data.
- Why: the same brand in the same list layout was being sent back for a hand answer again and again.
- Numbers that moved: Profound in Perplexity went from 149 to 153 of 400 answers, and in Google AI Mode from 10 to 11 of 25.
-
16 October 2026, 14:28 India time: 26 waiting hits answered, and patterns learned only from list-like layouts
- What changed: the 26 hits waiting for a person were answered (20 yes, 6 no). A yes now teaches a pattern only when the brand starts a list item, a bold line or a heading, sits in a table cell, or follows a label ending in a colon; a yes on an ordinary sentence start teaches nothing.
- Why: to clear the hits waiting for a person, and so that a later sentence that merely begins "Profound..." can never be counted by accident.
- Numbers that moved:
Brand Engine Before After Profound ChatGPT (750 answers) 435 (58%) 439 (59%) Profound Perplexity (400 answers) 153 (38%) 161 (40%) Profound Google AI Mode (25 answers) 11 (44%) 13 (52%) Keyword Insights Perplexity 11 16 Conductor ChatGPT 21 22
The order of the 15 brands named most across the engines did not change. In Google AI Mode alone, Profound (13) moved ahead of Otterly (12). Seven per-question top-3 lines changed (questions 8, 9, 12, 13, 16 and 21), for example Keyword Insights entering Perplexity's top 3 for question 21.
-
16 October 2026, 14:28 India time: Semrush set to "acquired, live"
- What changed: Semrush's status became "acquired, live" with owner Adobe (announced 19 November 2025, completed 28 April 2026), and our written summary gained a line saying the top brand on every surface is owned by Adobe.
- Why: Adobe's and Semrush's own newsrooms confirm the purchase closed, and Semrush still runs under its own name.
- Numbers that moved: no counts. Only Semrush's status and the tally of statuses change (126 live and 1 "acquired, live" became 125 and 2). Semrush stays first by mention rate on all four surfaces (ChatGPT 78%, Perplexity 88%, AI Overviews 75%, AI Mode 76%) and stays in every ranking; re-running the count with Semrush set back to "live" gives the same numbers.
-
16 October 2026, 18:45 India time: Scorecard weights set, and the public 60 renamed "openness score"
- What changed: the first scorecards, written at 18:07 the same day, used a proposed split of the public 60, because the published weights could not be found: price and plan limits 10, engines 10, runs and sampling 10, model and surface 10, freshness 5, exports and API 10, free tier 5. They now use the set weights: coverage 15, transparency 15, price 15, freshness 10, exports 5, with accuracy 30 and setup time 10 pending access. Transparency averages runs and sampling with model and surface; price averages price and plan limits with the free tier. The public 60 is now called the openness score everywhere.
- Why: the weights were set on 16 October. The new name says what the score is: it rates what a vendor publishes about its product, not how good the product is.
- Numbers that moved: 62 of the 64 openness scores (CitedSpy 60 and Evertune 52 are unchanged). The mean across the 64 tools still live after this change (XFunnel left out of both) went from 44.4 to 46.2; over all 65 tools in the first scorecards it was 44.1.
- Top: CitedSpy 60, Gumshoe 58, Elmo 56, Rankscale 56 became CitedSpy 60, Elmo 58, Gumshoe 57, then nine tools tied at 56.3 (Ahrefs, AI Peekaboo, AthenaHQ, Gauge, OpenLens, Otterly, Scrunch, SE Ranking, Trakkr).
- Bottom: Brandi AI 17, Brandlight 18, Bluefish 20, Adobe Brand Visibility 21 became Brandlight 21.8, Adobe Brand Visibility 23.3, Brandi AI 24.3, Bluefish 25.8.
- The five study brands: Semrush 42 to 44.5, Ahrefs 55 to 56.3, Otterly 55 to 56.3, Profound 53 to 53.3, Peec AI 49 to 47.3.
- Largest rises: CITED 27 to 34.8, Brandi AI 17 to 24.3, RivalScope 31 to 37.8. Largest falls: Wellows 47 to 44.3, Rankscale 56 to 54, Peec AI 49 to 47.3.
-
16 October 2026, 18:45 India time: XFunnel set to "acquired, dormant"
- What changed: new status "acquired, dormant" (bought, site still up but no longer sold on its own), first used for XFunnel, owner HubSpot. Recorded with the other ownership checks.
- Why: HubSpot's newsroom announced its agreement to acquire XFunnel on 31 October 2025; xfunnel.ai is dormant and HubSpot now sells HubSpot AEO instead.
- Numbers that moved: live AI visibility tools went from 65 to 64, and the dictionary's live brands from 125 to 124 (the 2 "acquired, live" brands unchanged). XFunnel left the openness table (its score was 25 under the proposed weights, 29.5 under the set ones). No answer count moved: XFunnel was named in none of the answers. The free checker counts (52 tools with one, 155 free tools, 28 with no account or email, 37 run, 19 results) did not change; XFunnel had no free checker, so the list of live tools without one went from 13 to 12, and the free checker line now reads "52 of the 64 live tools".
-
16 October 2026, 18:45 India time: Control verdicts from the checkers' own prompts
- What changed: when a free checker does not put the control (SEO Informatica) low, the prompts it showed for the control decide what that means, in a new control verdict: "mistake" if half or more of the prompts it showed for the control were about AI SEO or AI visibility tools, "different category" if fewer than half were, "unverifiable" if it showed no prompts. Agency questions count as a different category, because our 25 questions ask about tools. The prompts themselves are recorded word for word, in the free checker prompts download on the data page (667 prompts, shown by 11 of the 19 checkers that returned a result; the other 8 showed none).
- Why: a high score for the control is only a mistake if the checker asked the question we measured.
- Numbers that moved: no scores changed. The control paragraph changed from one list of 7 checkers giving the control a score that is not low (Otterly 23, HubSpot 42, Botric 83, Foglift 100, Bloomiro 83, Cituna high, Beamtrace 20) to: mistake 0, different category 2 (Foglift, Beamtrace), unverifiable 5 (Otterly, HubSpot, Botric, Bloomiro, Cituna).
-
16 October 2026, 18:45 India time: Bot-checked checkers read "needs a human click"
- What changed: the 8 checkers that stopped at a "verify you are human" check (Semrush, Ahrefs, Indexly, OpenLens, VisibilityTool.ai, Searcherries, SEOJuice, Parse.gl) are labelled "needs a human click" in the checker results (instead of "blocked bot check") and in the list of checkers (instead of "unclear"), and are left for a person to run by hand. They are never worked around.
- Why: to record that a person can still run these checkers by hand, which the scope notes now say, and that the checks are never worked around.
- Numbers that moved: none.
-
16 October 2026, 20:47 India time: Self-naming prompts are not a visibility measurement
- What changed: a checker prompt that contains the brand being scored, in any spelling (for example "What is Seoinformatica?"), is "self-naming", and the checker's result on it reads "not a visibility measurement". Where a checker shows every prompt and a result for each, the self-naming ones are removed and its score is recomputed on the rest; where every prompt it showed is self-naming, the whole score reads "not a visibility measurement"; where it gives one score over several prompts, the row is flagged. Self-naming questions are left out of the control verdict.
- Why: an answer to a question that names the brand will talk about the brand, so it cannot show whether buyers see it.
- Numbers that moved: 29 of the 667 prompts are self-naming (Trakkr 7, MaxAEO 8, Beamtrace 6, Foglift 6, Gauge 2).
- Foglift: all six scores (100 each) now read "not a visibility measurement", so it left the agreement tests. Before, it was listed as disagreeing (mean gap 54 points), and its four gaps of 40 points or more left that list (Otterly +52, Profound +60, Peec AI +66, SEO Informatica +100).
- Trakkr: Semrush went from 86 to 83 (named in 5 of 6 questions instead of 6 of 7); the other brands did not change. Trakkr's mean gap went from 32 to 31 points.
- Gauge: Semrush from 161 to 160 topics, Peec AI from 27 to 26.
- MaxAEO: Peec AI from 8 to 2 keywords. MaxAEO on Profound (2 of the 20 keywords it shows) and Beamtrace (1 of 5 prompts per brand) are flagged, their scores unchanged.
- Control: Foglift moved from "different category" to "not a visibility measurement"; the count of 7 checkers giving the control a score that is not low stayed the same.
-
16 October 2026, 20:47 India time: Trakkr compared with ChatGPT, not Google AI Mode
- What changed: a bug fix. The engine label "OpenAI model" contains the letters "ai mode", so the match for Google AI Mode caught it, and Trakkr's results had been set against our Google AI Mode rates since the first checker comparison, at 18:07 the same day. The Google AI Mode pattern now matches only the whole words "AI Mode", so Trakkr is set against our ChatGPT rates, the study's OpenAI engine (Trakkr names only an "OpenAI model", model not named, web search off). No other checker that returned a result was affected.
- Why: each checker must be compared with the engine it says it asked.
- Numbers that moved: Trakkr's mean gap went from 31 to 36 points; together with the self-naming change made at the same time, the published figure went from 32 to 36.
- Otterly: gap from -48 to -62 (set against 62% instead of 48%).
- Profound: from -52 to -59.
- Peec AI: from -36 to -53, which puts it on the list of gaps of 40 points or more.
- Ahrefs: from +44 to +38, which takes it off that list.
- Semrush: from +7 to +5. SEO Informatica: 0 either way.
- Knowatoa, which returned no result, had the same bug (its label "on-device AI model" contains "ai mode"): its Semrush row is now set against the mean of the 4 measured engines, 79%, instead of Google AI Mode, 76%.
-
19 October 2026: A page on Google AI Mode added
- What changed: a new edition page, which tools track Google AI Mode, sets the tools that list Google AI Mode among the engines they track beside the tools that AI Mode itself named in our answers.
- Why: the edition already held both halves of that answer: the engines each vendor lists, and what AI Mode named.
- Numbers that moved: none. The page reads the same frozen data, and no published number or rule changed.
-
6 November 2026: Gauge's plan and pages read again
- What changed: Gauge's Growth plan now tracks 3 AI platforms (ChatGPT, Google AI Mode and Microsoft Copilot) instead of 6, with 3 seats instead of 5 and 54,000 answers a month instead of 108,000, at the same price. Google AI Overviews, Gemini, Perplexity and exports moved to Enterprise, and Gauge's documentation now covers a public API for its AI visibility product. Gauge's page and the tool tables show the new reading, with its date, in place of the edition's.
- Why: on 5 November 2026 we were told that Gauge's pricing page no longer matched our reading of 16 October, so we read its pricing page and documentation again. A dated reading should say what the vendor's own pages say.
- Numbers that moved: only Gauge's plan facts, with its for and not-for lines worded to match. Every level behind its openness score held, so its score stays 56.3 of 60 and its place in each list is unchanged. No measured rate and no other tool moved. The data downloads keep the edition's reading of 16 October.
An earlier protocol, before the study ran
Edition 1 of this benchmark superseded the protocol below, which was never run.
A prepared monitoring trial, with no results claimed. Our dated trial protocol fixes four public buyer questions about our own site, one requested US English desktop Google AI Mode context, and two collection days. It separates setup time, waiting, review effort and the evidence actually retained. It also discloses our interest in the business being observed.
The packet is not run. The CSV observation worksheet has planned questions and blank results; the JSON method packet keeps the conditions, sources and missing observations explicit. Confirm compatible access and freeze the actual dates before collection. A missing answer is not a negative mention, and different independently generated answers do not by themselves show that a tool is wrong.
Access can prevent a fair trial. For example, Profound’s published Trial terms, reviewed 7 October 2026, do not allow custom prompts or exports. That plan cannot complete this particular task; it is not a finding about paid-product quality.
For a larger rollout, prepare a deployment scope packet. Before committing to the spend, use the purchase decision guide to check readiness and the contribution required to recover it.
The Clarity Audit
Want to know who AI names in your market?
The Clarity Audit is free. We look at your website and your market, show which businesses Google and AI answers name for the searches your buyers make, and list the three fixes to consider first. No call needed.