What ChatGPT and Claude searched in 160 tracked answers

We looked at a week of Reputably’s own AI results to work out what our website needed to explain better. Across 160 answers, 33 showed the searches behind them. Those answers contained 69 searches, surfaced 97 domains and cited 57. The repeated buying questions helped us decide which comparison and workflow pages to improve.

The review covered 9 September 2026 at 13:43:41 through 16 September 2026 at 13:43:41 AEST (03:43:41 UTC at both boundaries), using the recorded runs from 10, 14, 15 and 16 September.

What we checked

We used the stored results for our own business. The two main questions were “best social listening tools for leads” and “best ai rank tracking tools.” Each had four wording slots, but two slots in the AI-ranking group repeated the same text. That left seven distinct questions. Some of the wording also asked about traditional keyword-ranking software.

There were 32 records each for ChatGPT, Claude, Gemini, Google AI Mode and Google AI Overview. ChatGPT, Gemini and the Google surfaces were recorded through DataForSEO; Claude records were labelled direct. Reputably’s documented Claude collection uses its plan-based web-search path. These records are not recordings of personalised consumer sessions.

Which answers showed their searches

Engine	Answer records	With disclosed searches	Search occurrences
ChatGPT	32	1	5
Claude	32	32	64
Gemini	32	0	Not disclosed
Google AI Mode	32	0	Not disclosed
Google AI Overview	32	0	Not disclosed
Total	160	33	69 recorded

Some answers did not show their searches. We left those search counts unknown. The 69 searches we could see used 32 distinct phrases; these are the engines’ search phrases, separate from the seven questions we asked.

What the repeated searches asked for

The repeated phrases were about choosing: features, comparisons, reviews and which tool suits the job. That helped us choose a buyer’s guide to work on. Because we selected the original questions, these results describe our test rather than the wider demand for those topics.

The one ChatGPT answer with visible searches included queries limited to Brandwatch, Sprout Social, Meltwater and Talkwalker. A Reputably page could not appear in those searches. We took that as a reason to investigate product coverage on other sites as well as improve our own pages.

Why 57 cited sites and 530 cited URLs are both correct

Within the 33 records that disclosed searches, we counted 97 surfaced domains from linked query hits plus separately recorded browsed pages, and 57 cited domains from those records’ final citation lists. Domain counts remove a leading “www.” and preserve other subdomains.

Looking at all 160 answers, including the ones that did not show searches, we counted 1,130 citation occurrences and 530 different URL strings. Those URLs include tracking parameters, and several can belong to the same website. That is why this total is larger than the 57 cited domains in the smaller group.

About the analysis

This article reports a dated review of our own tracker. The original records remain private. We counted distinct answer records, exact search phrases and exact URLs, except where a figure explicitly counts domains. This article is not a live dashboard.

What we did with the findings

The review led us to work on a social-listening comparison for people looking for leads, clearer explanations in our AI visibility comparison, and a guide to using these searches when editing content. The aim was to answer the buying questions we kept seeing. Google’s advice on helpful content with original value is a useful starting point for that work.

Google describes how its AI search features can run related searches. Our records did not expose those Google searches, so we cannot reconstruct them here. We also cannot use this week of results to predict when a new page will rank. The next step is to make useful changes and keep checking the same questions.