What four AIs did when we asked them 180 buying questions

We asked ChatGPT, Claude, Gemini and Perplexity the same 180 buying questions, and they did four different things. ChatGPT found the most pages and cited the fewest. Claude named the most brands. Gemini ran no web search on 73 of its 180 answers. And where all four named a brand, they named the same one first on only 5 of 126 questions.

These are findings from Citewake's Edition 0, published 29 October 2026. The full reports are at /editions/2026-10.

What exactly did we ask?

Edition 0 covers three categories, each with 60 scored questions that name no brand:

• AI search visibility tools
• Link-in-bio tools
• Live race tracking

That is 180 questions. Each was asked of four assistants, through their official interfaces, on 28 October 2026. The models were ChatGPT (gpt-6-luna), Claude (claude-haiku-5-5), Gemini (gemini-3.1-flash-lite) and Perplexity (perplexity/sonar). The request asked each to search the web before answering. In all, 720 answers were planned and 712 were valid; the eight that failed were all in live race tracking.

No location was sent, so the results describe a location-unspecified market, not the world. The reports mark Edition 0 as incomplete: cited-page presence was not measured. Every other measure is complete. The method is on the methodology page.

Did all four AIs search the web?

No, and this is the sharpest difference.

ChatGPT and Perplexity searched on every answer in all three categories. Claude did too, except in link-in-bio tools, where it searched on 98.3% of answers. Gemini searched on 61.7%, 53.3% and 63.3% of its answers across the three categories. Taken together, Gemini ran no web search on 73 of 180 questions, which is 41%, even though it was asked to search.

This matters because an answer without search comes from the model's memory. See when an AI answers without searching.

How many pages did they find, and how many did they cite?

The gap between finding and citing is large for some assistants and nonexistent for another.

• ChatGPT found a median of 43.5, 32 and 33 pages per answer across the three categories, and cited a median of 3, 2 and 2.
• Claude found a median of 18, 17 and 18, and cited 7, 6 and 7.
• Perplexity found a median of 15 in each category and cited 5, 3 and 4.
• Gemini reported the same median number of sources found and cited in every category: 12, 7 and 9.

So the sources an answer shows can be a small part of what the assistant looked at. The concept is explained in pages found vs. pages cited. Each provider reports different evidence, so these counts are not strictly like for like.

How many brands did each name?

This is where the styles split.

Claude named a median of 7.5, 7 and 7.5 brands per answer in the three categories. ChatGPT named 3, 2 and 2, and Perplexity 3, 2 and 2. Gemini sat between, at 6, 4.5 and 6.

The answers differed in length too. Claude's median answer ran to 433, 341 and 415.5 words, against ChatGPT's 208, 116.5 and 149.5.

A brand therefore has more room to be named by Claude than by ChatGPT or Perplexity, before anything about quality is considered. This is part of why AI assistants disagree.

Did they agree on the best product?

Hardly. In the three categories there were 43, 39 and 44 questions where all four assistants named a brand. That is 126 in total. On how many did all four name the same brand first? One, two and two: 5 of 126.

First position is not always a preference, so this is a strict test of agreement, not of quality. But it supports a simple point: there is no stable "number one" that all assistants share. See why there is no real rank #1 in AI answers.

What did each category's leaders look like?

The reports give each brand a recommendation rate with a 90% interval, overall and per assistant.

• AI search visibility tools: the most-recommended brand was Otterly.AI, at 41.67% (interval 36.34% to 46.93%), followed by Peec AI and Profound.
• Link-in-bio tools: Linktree led at 56.68% (48.71% to 64.19%), followed by Beacons and Lnk.Bio.
• Live race tracking: RaceJoy led at 26.23% (20.19% to 32.41%), followed by Racemap and Garmin LiveTrack.

How to read these ranges is in measuring recommendation rate with confidence intervals.

How much did one brand vary between assistants?

A great deal.

• Otterly.AI was recommended in 16.67% of ChatGPT's answers (interval 8.97% to 25.00%) and 66.44% of Claude's (55.24% to 76.78%).
• Racemap was recommended in 1.85% of ChatGPT's answers (0% to 5.26%) and 29.19% of Claude's (19.44% to 38.82%).
• Linktree was recommended in 66.98% of ChatGPT's answers and 49.12% of Perplexity's.

An overall rate hides this. A brand strong with one assistant can be weak with another, which is why the reports show the per-assistant rates.

Which sites did the assistants cite?

The cited-domain lists differ by category and by assistant.

• In AI search visibility tools, semrush.com was cited in 42 of the 240 answers (17.5%). Reddit was cited in 26 of Gemini's 60 answers (43.3%), more than any other domain for Gemini.
• In link-in-bio tools, linktr.ee was cited in 74 answers (30.8%), including 42 of ChatGPT's 60 (70%).
• In live race tracking, apps.apple.com led overall, in 28 answers (11.7%), and in 16 of Claude's 60 (26.7%). Reddit was cited in 18 of Gemini's 60 (30%).

For the kinds of sites behind these, see which kinds of websites AI assistants rely on.

What should you take from it?

Four things, each within what the data supports.

1. Treat each assistant separately. They search, cite and name brands in different ways.
2. Do not read one answer as the category's verdict. The first pick rarely matches.
3. Check whether an answer searched. Memory answers favour what was already known.
4. Keep measuring. Edition 0 is a first look, one moment in time, not a trend. See how AI recommendations change over time.

Every number here comes from the Edition 0 reports, which also publish the answers behind them. Start at /editions/2026-10. To see how assistants describe your own site today, try the Citewake checker.