We asked the AI engines the same 36 buyer questions two months later

What changed when we asked the AI engines the same 36 questions again?

On October 26, 2026 we put the same 36 buyer questions to ChatGPT, Claude and Perplexity that we first asked on August 24, with the same models, the same wording and web search on. GrowthMasters was cited in 4 of the 108 answers, up from 1. All four answers cite pages we published on our own site in August and September. Claude cited us for the first time. ChatGPT cited us in neither run.

The engines moved more than we did. ChatGPT searched the web on 6 of the 36 questions, down from 11. Claude searched on 16, up from 11. Perplexity still cited sources on all 36, but about half as many sites per answer, and LinkedIn, the site it cited most in August, fell from 28 of its answers to 5.

We said in our October newsletter that this re-run would happen on October 24. It ran two days late, on October 26, and every October number here comes from that run.

definitionAnswer audit

An answer audit puts the questions a company's buyers ask to the AI engines, with web search on, and records which sites each answer cites. Run again with the same questions and models, it shows what moved, which a single run cannot.

Which GrowthMasters pages got cited, and on which questions?

Four pages, on three questions.

• Asked whether a company should fix its CRM data before or after building AI go-to-market workflows, Perplexity cited Should you fix your CRM before building AI GTM workflows?
• Asked how to hand off an AI go-to-market system to an internal team so it survives, Perplexity cited How we build a go-to-market AI employee.
• Asked what AI GTM maturity is and how it is measured, Perplexity and Claude both cited The AI GTM stack that looks Stage 4 and runs at Stage 2, and Claude also cited our verification criteria.

In August our one cited answer was on that last question too, from Perplexity, which cited our maturity model page and a LinkedIn post by one of our cofounders. Neither of those two was cited in October.

A citation is not a recommendation. We searched the text of all 108 answers and none of them named GrowthMasters. The engines used our pages as sources and answered in their own words.

One of the three posts, the CRM post, carries a close version of the buyer's question as its title. The other two answer their question inside the post under different headings. So a matching title is not required to be cited, and one re-run cannot say whether it helps.

Why did LinkedIn drop in Perplexity's answers?

We do not know, and Perplexity does not publish how it chooses sources. Here is what changed.

In August, 28 of Perplexity's 36 answers cited linkedin.com, and those answers pointed at 37 individual LinkedIn posts, 35 of them published from a person's profile. In October, 5 answers cited linkedin.com, and only 2 of those citations were individual posts: one from a person and one from a company page. None of the 37 August posts was cited again. Reddit, the second most-cited site in August with 18 answers, was cited in none.

Part of the drop is that Perplexity now cites fewer sites overall. Across its 36 answers in our October run it made 331 site citations, counting each site once per answer, down from 616, about 9 per answer instead of 17. LinkedIn fell further than that: its answers dropped by 82 percent while Perplexity's total citations dropped by 46 percent.

This matters for two of our own posts. The 37 LinkedIn posts an AI engine cited and The founder posts, the company page keeps the record were built on the August LinkedIn pattern. The second said it would carry a correction if company pages entered the cited set by the re-run. A new company page did, in a set that shrank from 37 posts to 2, and both posts now carry a note with these numbers. The August finding described what one engine cited on one day, and two months later that engine cited something else.

How did ChatGPT and Claude change?

Both decide for themselves whether to search the web, even with search enabled, and in both runs every answer that searched cited sources and every answer that did not cited nothing. So their citation rates moved exactly with how often they chose to search.

ChatGPT stopped searching on five questions and started on none, including whether to build or buy an AI outbound system and whether a startup should hire a VP of Sales or a growth agency first. Claude started searching on five and stopped on none, including how to build a signal-based outbound system and the difference between AI-assisted and AI-orchestrated go-to-market. On two questions they swapped: whether to use one full-service agency or several specialist vendors, and how to tell whether a company is behind its competitors on AI. Across all three engines, 58 of the 108 answers cited sources in August and 58 did in October, so the total held while the two engines moved in opposite directions.

What does every answer look like, August against October?

What can a re-run like this not show?

It cannot show a trend. Each question went to each engine once in August and once in October, and engine answers vary from one run to the next, so a single question flipping can be noise. It does not measure how many people see these answers, or traffic, leads or revenue. It cannot say that our pages caused the four citations, only that they were cited. And it cannot show what the engines would have answered on October 24, the date we gave.

What happens next?

We will put the same 36 questions to the same three engines again in January and publish what moved, including the questions where we were cited and the ones where we were not. The raw responses from both runs are kept, so the comparison can be checked against them.