On August 24, Perplexity cited a list of the best SEO and AEO companies for law firms, published by one of the brands our tracker follows, in all five answers it gave to "AEO agency for law firms." On September 13 it cited that page in none of twelve answers to the same question. On October 4, none of twelve again.
Data as of October 2026: every source Perplexity cited for four law firm marketing questions in our citation tracker, collected on August 24 (5 runs per question), September 13 and October 4 (12 runs each), through the same API endpoint and model setting.
The easy reading is that the page slipped, or that a competitor outworked its owner in three weeks. Both readings assume Perplexity keeps a ranked list for each question and moves pages up and down it, the way a page of search results does. If that were true, most of the August list would still be there in September, with a few pages traded at the edges.
So we went back and counted what it kept.
Twelve identical answers in one sitting
Start with the part that looks stable. On October 4, each of the four questions got exactly the same source list in all 12 runs. On September 13 the most common list showed up in between 5 and 11 of the 12 runs, depending on the question, and on August 24 in 4 or 5 runs out of 5. A separate check we ran on October 21, ten Houston lawyer questions asked ten times each, came back with a single source list per question every time.
Within a day, little moves.
That stability is less useful than it looks. All the repeats on a given date went out within hours of each other, and we cannot rule out a cache serving the same retrieval to every call in that window, a problem we lay out in how to measure an AI citation rate. Either way, asking a question twelve times in one afternoon tells you what Perplexity says today and next to nothing about what it will say next month, because the movement we found happens between dates and hardly at all between runs.
Three weeks later, four in ten were gone
August 24 to September 13 is the cleanest comparison we have. Same endpoint, same model setting, the same 20 source slots per answer, three weeks apart.
| Question asked on Perplexity | Domains cited Aug 24 | Still cited Sep 13 | New on Sep 13 |
|---|---|---|---|
| how do law firms show up in ChatGPT recommendations | 20 | 8 | 12 |
| best SEO and AEO companies for law firms | 18 | 16 | 10 |
| AEO agency for law firms | 19 | 8 | 12 |
| GEO agency for personal injury law firms | 19 | 13 | 7 |
Across the four questions, Perplexity dropped 31 of the 76 domains it had cited in August. Of the 86 domains it cited in September, 41 had not appeared three weeks earlier. That is 48% new.
The closest outside figure we know of comes from Profound, a company that sells AI visibility tracking, so read it as vendor research. In a July 2025 study of roughly 80,000 prompts per platform, Josh Blyskal and Sartaj Rajpal found that 40.5% of the domains Perplexity cited in July had not been cited for the same prompts in June, a monthly turnover not far from the 48% we saw over three weeks. Their verdict: "Citation drift isn't a bug in AI systems. It's an inherent feature of how these probabilistic models operate."
Our four questions are a thin sample next to theirs. September also had more runs than August, 12 against 5, which can only push our new count up a little, given how closely runs on the same day matched. Still, two very different samples landing that close points to turnover as the normal state of a Perplexity answer, not a bad month for one page. The rate also swung by question: "best SEO and AEO companies for law firms" kept 16 of its 18 domains, while "AEO agency for law firms" kept 8 of 19.
Why we stopped the clock at September
We would rather show October as well, but the instrument changed under us. Every one of the 384 Perplexity answers in our October run came back with exactly 10 sources. In September, 359 of 384 came back with exactly 20, and in August 140 of 155 did. Same endpoint, same model setting. Perplexity's API changelog for September and October announces a new Fast Search mode but says nothing about how many sources its Sonar models return, and we have not found a published explanation anywhere else.
Comparing October's list with September's would mostly measure that cap. What the cap does allow is a different question: when Perplexity suddenly had half as many slots to fill on every answer, which of the pages it already knew did it keep, and which did it let go?
Mostly, it kept incumbents. Of the domains in October's shorter lists, 26 of 39 had been cited in September, broken down as 5 of 10, 9 of 9, 4 of 10 and 8 of 10 across the four questions in the order of the table above. How many incumbents survived depended heavily on the question.
The pages that kept their place
Fourteen pages were cited at the same URL on all three dates, for at least one of the four questions. Judging by their URLs, seven are lists of agencies: the list from the opening, two pages on The National Law Review's site, and pages on 6smarketers.com, bluelinks.agency, gtm8020.com and posirank.com. Three are agencies' own service pages, on firstpagesage.com, fuelonline.com and toppeconsulting.com, and three are explainers, on resultfirst.com, lseo.com and LinkedIn. The fourteenth is a page on spotfund.com that we could not place from its URL alone.
Search engines other than Perplexity show the same shape. SISTRIX reported in April 2026, in a study of 82,619 prompts over 17 weeks that we have only seen summarized, that 86% of prompts in Google's AI Mode had a stable core of one to five domains.
No press release survived. On "AEO agency for law firms," Perplexity cited three syndicated press releases on August 24, two on September 13 and three on October 4, on hosts that included nbcrightnow.com, openpr.com, digitaljournal.com and theglobeandmail.com. One release carried over from August to September. None of October's three had been cited before. On this evidence a syndicated release buys a few weeks in the answer, and then newer releases take the slots.
We did not track whether any of these pages changed between dates, so we cannot tell whether the survivors were refreshed while the dropped pages sat still, and that is the first thing we would check with another quarter of data.
One page, kept on one question and dropped from another
The list from the opening is the cleanest case we have, because it is one URL with one history. On "best SEO and AEO companies for law firms," Perplexity cited it in 5 of 5 runs on August 24, 9 of 12 on September 13 and 12 of 12 on October 4. On "AEO agency for law firms" it went from 5 of 5 to none, and stayed there.
Nothing about one page can explain both histories. What changed was the kind of source Perplexity reached for. On August 24, judging by URLs, 10 of the 20 pages it cited for "AEO agency for law firms" were lists of agencies, 3 were agencies' own service pages, 3 were press releases and the other 4 were guides to choosing or pricing an agency. On October 4, 2 of 10 were lists and 5 were service pages, with the other 3 press releases.
Our best guess is that Perplexity started reading "AEO agency for law firms" as a request for an agency, where it used to read it as a request to compare agencies, and a list stopped being the natural answer. We cannot test that from outside. If you track a question where Perplexity's sources swung from lists to providers, or the other way, we would like to compare notes. Two cases beat one.
What a three-week citation is worth
If someone shows you your Perplexity citations from one day of runs, ask for a second date at least three weeks later before you act on it, and ask whether the number of sources per answer stayed the same in between. One day is one observation.
If you are deciding where to spend, the pages that held their place across our three dates were mostly lists of agencies, and none was a press release, so a place in a list Perplexity keeps returning to is worth more than a release it will cycle out within weeks. For law firms, our breakdown of how law firms show up in ChatGPT recommendations names the directories and publishers the engines lean on for that question. And if you want your own firm tracked across dates this way, with the source cap and the same-day repeats accounted for, that is what our AI visibility audit does.
Matthew (Matt) Bertram helps energy and industrial companies get found, trusted, and chosen as AI reshapes how buyers decide. As owner and CEO of EWR Digital, he leads the agency's AI search and SEO work for oil and gas and other capital-intensive industries; he is also President of ModalPoint. He is also Chief Marketing Officer of the Oil & Gas Global Network (OGGN) and the author of multiple books, including LLM Visibility: A Decision-Grade System for Winning AI-Mediated Discovery and the co-authored Oil & Gas Sales & Marketing: The Energy Growth Playbook for Oil and Gas Leaders. He is a member of the American Petroleum Institute's Houston Chapter and the International Association of Privacy Professionals (IAPP).


