Four weekly runs, 576 AI answers: askedabout.com was cited in 18 of them, from 4 of 48 engine×question cells and exactly 2 URLs; the one cell that holds is a ranked list on a page Google ranks 79th

Method, stated before the numbers

Every Monday at about 13:07 UTC our audit engine asks the same 12 buyer questions about AI visibility to ChatGPT, Perplexity, Gemini and Claude through their search-grounded APIs, three times each, and stores the full answer plus the citation list each engine returns. That is 144 answers per run and 48 engine×question cells of three samples each. A sample counts as citing us only when the hostname askedabout.com is in the engine's structured citation array. A prose mention of the brand does not count; a bare URL typed into the answer text does not count. Rank is the 1-based position of our URL in the engine's own ordered list, and "sources" is the number of distinct domains in that list.

Four scoring runs are on the ledger: 2026-08-03, 08-10, 08-17 and 08-24. The run before them, 07-28, was the verification run for a sensor fix; it read 0 of 144 with a positive control passing 11 of 11, so the 08-03 zero below is a reading, not a dark instrument. The question shapes in the tables are hand labels we assigned to the 12 questions before cutting the data: "tool list" (which tools…), "alternative list" (a good X alternative…), "how-to" (how do I…), "price" and "offer/price" (cheapest… / one-time instead of subscription…), and one vertical tool-list question (dentist or law firm).

The series: 0, 3, 7, 8

Two engines account for every citation and two have never cited us on any question in any run. The 18 are split exactly 9 and 9 between ChatGPT and Perplexity, but they are different kinds of citation. Every one of ChatGPT's nine points at the homepage, and in all nine the answer also names AskedAbout in its text. Every one of Perplexity's nine points at one deep page, and in none of the nine does the answer text mention the domain at all: it is a source-list citation and nothing else.

The four cells, run by run

One cell is stable and three are not. Perplexity on the Profound-alternative question has cited the same page in every one of its nine samples since the page entered the corpus, and the page's rank inside Perplexity's list has moved from third to fifth to second while the number of domains competing on that question grew from 41 to 53 to 58 to 73. The three ChatGPT cells are all homepage citations and they move every week: q34 went 3 of 3 to 1 of 3, q42 went 1 of 3 to 3 of 3, and q35 appeared for the first time on 08-24 with a single sample. We built a ranked-list page for q35 on 08-17; ChatGPT cited the homepage on that question, not the list, so for the list's own pre-committed read it is uncited.

The 44 cells that never cited us, by question shape

The control: same question shape, different page shape

The two alternative-list questions are the cleanest comparison in the panel because they are the same sentence with a different vendor name, and we hold different kinds of page for each. For the Profound question we publish both a head-to-head page ("AskedAbout vs Profound") and a multi-vendor ranked list with prices ("The best Profound alternatives for a small business", published 2026-07-31). For the AthenaHQ question we publish only the head-to-head ("AskedAbout vs AthenaHQ"). Over four runs the ranked list was cited in 9 of the 9 Perplexity samples since it entered the corpus; the two head-to-head pages were cited in 0 of 576 answers on any question.

Google's ranking does not order these pages the way citation does. The head-to-head page for the Profound question ranks 19 places higher on Google than the ranked list and has never been cited; the list ranks 79th and is the most reliably cited page we have. That is consistent with what we found when we first cut this question a month ago: the pages Perplexity cites on it are almost all multi-vendor lists on domains our size or smaller, and access was not the difference (our head-to-head page was indexed, crawlable by every AI crawler and topically exact). What this cut adds is the within-site control, one question shape and two page shapes, and it comes out the same way three weeks running.

The pricing table: 0 of 96

The other prediction we made from the list-shape reading was that a ranked price table would be cited on the two pricing questions. We built one on 2026-08-01: seventeen vendors' entry tiers read from their own pricing pages, billing labels quoted, with the only one-time purchase in the table being ours. In four runs it has been cited 0 times in 96 answers, on all four engines, while those two questions cite 44 to 63 distinct domains per run. The page shape that works on q39 does nothing on q37 and q38. Whatever gate is being applied, it is applied per question, and a shape that passes it on one question does not pass it on another.

How many sources each engine returns

The two zeros are not the same size. Claude returned no citations at all in 47 of its 144 answers, so its 0 is measured on 97 answers that carried any source; Gemini's is on 129. Perplexity returns the widest lists and is the engine where a deep page got in; ChatGPT returns short lists and is the engine where only the homepage got in. A rank of 2 among 20 Perplexity sources and a rank of 4 among 8 ChatGPT sources are not comparable positions, which is one reason we report both numbers rather than a share.

Limits

• This is one domain's history, not a rate. Eighteen citations from four cells is enough to say which of our pages the engines pick up and which they do not; it is not enough to generalise to any other site.
• Three samples per cell per run. A 3 of 3 held on three runs is nine draws; a 1 of 3 is one draw and can be noise. We report the single-sample cells because leaving them out would understate ChatGPT's movement, not because they mean much on their own.
• The shape labels are ours, assigned to the 12 fixed questions before the cut. Another reader could label q37 as a how-to or q44 as an alternative list; the per-question rows are in the tables so the labels can be discarded.
• The homepage citations on ChatGPT coincide with a large volume of ChatGPT-User fetches of our homepage in the same weeks. This cut does not test whether retrieval traffic causes citation; it only records that the page cited is the page fetched.
• Our page inventory is uneven by design: we hold a ranked list for two of the twelve questions (q39, q35), a price table for two (q37, q38), head-to-head pages for two (q39, q40) and no dedicated page for the other six. Zero cells on questions we never built for are not evidence about page shape.

Pre-committed read: the 2026-08-31 run, read on 09-01

Three claims, each with the number that falsifies it, written before the fifth run. First: Perplexity cites /compare/best-profound-alternatives in at least 2 of 3 samples on the Profound-alternative question again on 08-31; a reading of 1 of 3 or lower falsifies "the cell is stable". Second: the AthenaHQ-alternative question stays at 0 of 12 across the four engines while we hold only a head-to-head page; a citation of that head-to-head page falsifies the list-versus-head-to-head split. Third: the two pricing questions stay at 0 of 24; a single citation of the price table falsifies "the gate is per-question" as stated here. We will publish the fifth run's cells either way.

What to do with this if you chase AI citations

• Build the page shape the engine already cites on that exact question, and check per question. On our panel a ranked list with prices was cited nine times out of nine on the one question it matched and zero times on the two pricing questions it was also built for. Read the cited set for your question before you decide what to write.
• Do not read Google position as a proxy. Our most-cited page ranks 79th; the same-topic page that ranks 60th has never been cited. If your AI-visibility plan is a Google ranking plan with a new name, the data on this site says it is measuring the wrong gate.
• Count per engine, and count answers with no citations at all. A "0 on Claude" is measured on two-thirds as many answers as a "0 on ChatGPT", and a rank-2 citation in a 20-source Perplexity list is a different thing from a rank-4 in an 8-source ChatGPT list. Report the denominator with the number.

The same cell-by-cell read, on your own domain and your own buyer questions, is what the free check runs. The ranked list of the tools the engines name on the question that asks it directly is here.