What the answer engines read across every software category
One question per category, put to the two AI engines that hand back the pages they opened.
The week of Sep 13, 2026
asked between Sep 18, 2026 and Sep 19, 2026 · 50 categories with a completed check · Perplexity and Google AI Overviews · US English
These are not the leaderboard's engines. The leaderboard and the debate ask ChatGPT, Claude and Gemini. This page asks Perplexity and Google, because they are the two that disclose what they read. What an engine reads here does not explain a rank there.
Perplexity · Most-read site
read in 29 of the 50 categories it answered
review site
Google AI Overviews · Most-read site
read in 16 of the 46 categories it answered
community
The two engines read different webs. Neither list below is the other's, and nothing on this page adds them together.
What each engine reads most
The sites each engine returned in more than one category, most categories first. These are two separate lists: a site high in one column may be absent from the other entirely, and no row here is a total across the two engines.
Perplexity
Perplexity retrieved 2,250 pages from 1,231 sites across the 50 categories it answered · 430 of those pages sit on a company's own site
29 of 5051review site
21 of 5025publisher
20 of 50126review site
17 of 5026publisher
13 of 5028publisher
12 of 5013publisher
12 of 5012publisher
11 of 5012publisher
8 of 5011publisher
8 of 5010publisher
8 of 509review site
8 of 509publisher
7 of 5012vendor site
7 of 5010review site
7 of 508publisher
These are the 50 most-read of the 202 sites read in two or more categories. 1,029 more sites were read in exactly one category.
Google AI Overviews
Google AI Overviews retrieved 378 pages from 263 sites across the 46 categories it answered · 75 of those pages sit on a company's own site
16 of 4619community
15 of 4623video
7 of 468publisher
4 of 464publisher
4 of 464community
4 of 464vendor site
3 of 463publisher
2 of 465vendor site
2 of 464vendor site
2 of 464publisher
2 of 463publisher
238 more sites were read in exactly one category.
Sites are counted by host and are never folded together: learn.g2.com and g2.com are two sites here, as they are on every category page.
What kind of site
Every page each engine retrieved, grouped by what kind of site it sits on. A vendor page is one on the site of a company some engine named or some leaderboard ranks; everything the lookup table does not recognise is a publisher.
Perplexity
- publisher1,611 pages
- vendor site430 pages
- review site195 pages
- community14 pages
- video0 pages
Google AI Overviews
- publisher240 pages
- vendor site75 pages
- review site16 pages
- community24 pages
- video23 pages
Named, by leaderboard rank
Every brand our leaderboards rank, grouped by where it ranks, and how many of them each engine's answer named. Each engine has its own denominator: a brand in a category where that engine was not measured is absent from its column, never counted as un-named.
This is a count of namings, not agreement and not accuracy. These are not the engines behind the rank: the leaderboard asks ChatGPT, Claude and Gemini, and what an engine named here neither confirms nor contradicts a rank there.
Where Google shows no AI Overview
In 4 categories Google's results page carried no AI Overview for the question. That is a measured property of those markets — the page was read and there was no overview on it — not a gap in this data and not a zero.
Google's first page, against what its AI Overview read
Google returns its classic results on the same call as the Overview, so the two lists can be compared directly. They turn out to be largely different lists — being on Google's first page for a question is not what gets a page read by the Overview above it.
123 of the 368 pages the Overview read across 45 categories are also on Google's first page for the same question
Per category: fewest 0, median 2, most 6
The page in the first position was also read by the Overview in 18 of 45 categories
Across those 45 categories, 91 ranked companies have their own site on Google's first page; its AI Overview named 45 of them and opened a page on the site of 32.
5 categories share no page at all between the two lists:
A check holds 7 to 10 classic results, so this is Google's first page and not a top 10 — and a position on it is a position of a page, never a company's rank.
Read is not recommended, across software
Every one of the 1,303 ranked brand pages across every category, by what the engines did with it there. "Named" means at least one engine's answer named it; "site opened" means at least one engine retrieved a page on the company's own site. They are different things, and the two middle rows are how different.
160
Named, and its own site opened
347
Named — its own site never opened
94
Its own site opened — not named
702
Neither named nor read
Counted by membership across both engines, so within a category no brand is counted twice and nothing here is a sum of the two engines. A company ranked in several categories has a row in each, judged on that category's own question. A brand in a category where an engine was not measured is judged only on the engine that was.
All categories
One row per category in this cycle, each linking to that category's full check — every page each engine retrieved and every company each one named.
Sites read by both — sites both engines read in this category. Present only where both answered: where one did not there is no list to intersect, and the dash is not a zero.
Named, not ranked — companies an engine's answer named that this category's leaderboard does not rank. A count only: the names are on the category's own page, where they were measured.
Also on Google's first page — pages Google's AI Overview read in this category that are also among Google's classic results for the same question. Present only where an Overview answered.
How this is measured
Once a week each category's existing leaderboard question is put to each engine on its own plain call, from the United States, in English. Every page an engine reports retrieving is recorded, and the companies its answer names are read back out of the answer text. Retrieved pages are not opened.
This page shows one complete cycle: the newest week in which every category has a settled check, or a week that has closed. It is never a mix of two weeks — a count that spanned two cycles would be a count of nothing anyone could cite, and every figure here is meant to be citable with its date. A closed week can hold fewer checks than there are categories; the number of checks it holds is printed above, and every count on this page is read against its own.
- Retrieved, never cited. The engines hand back the pages they pulled; neither says which of them it leaned on.
- Per engine, never pooled. Nothing on this page adds the two engines' pages, sites or categories together — the only cross-engine facts here are that both engines read a site and that neither named a brand.
- Counts, never rates. Every figure counts a named thing, and every "of N" is that engine's own N. There is no percentage anywhere on this page.
- An engine we could not read in a category is absent from that engine's counts on both sides. It is never shown as a zero.
These are not the leaderboard's engines. The leaderboard and the debate ask ChatGPT, Claude and Gemini. This page asks Perplexity and Google, because they are the two that disclose what they read. What an engine reads here does not explain a rank there.
Methodology versions 1.0.0, 1.1.0Read the full methodology →
Earlier weeks
This is the first complete cycle, so there is no earlier week to reproduce yet.
Citing this page
Respectarium. (2026-09-13). What AI Reads Across Software [AI Sources index]. https://respectarium.com/sources