Has AI published how it picks brands?
No. The official docs from OpenAI, Google and Perplexity cover how to allow crawlers and what makes a page eligible to appear; none of them has published scoring rules for which brand to recommend. Below we split things into two kinds: what can be seen directly in the data (observation), and what is drawn from the observations (inference). Both are labeled in the text.
Which web pages do the brands in an answer come from?
On 2026-10-26 we reviewed GammaSlide: we asked AI (using Claude Sonnet 5.5, with web search on) buyers' questions 21 times in total, and GammaSlide was mentioned 0 times. The brand recommended most in the answers was Gamma (12 times). AI read 28 cited web pages, of which 19 were pages listing multiple similar products, and 0 of them mentioned GammaSlide. The most-cited domains: creati.ai, slidesai.io, genai.works, ppt.ai.
Observation: most cited pages are third-party pages listing a batch of similar products, not each product's own website. Inference: when answering questions like “what products are there”, AI first looks for lists written by others, then picks the names that appear most often in those lists. So if you are not in those lists, it is hard to be in the answer.
What signals does AI look at when recommending brands?
| Signal | How it was observed | What you can do |
|---|---|---|
| Number of appearances in third-party rankings and comparison articles | The review above: 19 of the pages read were list pages | Get into real reviews and rankings, rather than writing a ranking yourself |
| Type of cited source | Cited domains are mainly review sites, blogs and tool directories | Look at which kinds of sites are cited for your questions, and fill in that kind |
| Whether official docs and your own pages are clear | Inference: AI reads the official site when it needs to check price and features | Write price, features and who it is for as text and tables that can be read directly |
| Ranking in search engines | Ahrefs study: only 12% of cited URLs were in Google's top 10 | Don't focus only on Google ranking; being cited and ranking are two different things |
| Stability across repeated asks of the same question | SparkToro study: when the same question is asked repeatedly, the recommended brands are not consistent | Sample many times and look at how often you appear; don't ask just once |
Sources of the numbers: Ahrefs, published 2025-08-11, 15,000 prompts; SparkToro × Gumshoe, 2026-01, 12 prompts each asked 60–100 times per platform. The two studies differ in sample and method, and may not apply to your industry.
What does a third-party review page look like when AI can cite it easily?
Pagent's “Email and Calendar” ranking page is an example we built ourselves. It puts each assistant's score on a fixed test, the number of public reviews and positive share, price and supported channels in one table, states an updated date on the page, and links to Pagent's methodology page explaining how it was tested. Inference: for AI to cite a review page, what it reads should ideally be dated, have a stated method, and have complete fields.
Pagent currently has 17 ranking pages covering 47 assistants, 22 of which have scores from our own testing; the rest are pre-filled initial scores. To avoid passing off pre-filled scores as measured ones, this article cites no specific score and uses only the page's structure as the example. Read on 2026-11-05.
How can a product team raise its chance of being recommended?
1. First find the pages cited for your questions
Ask AI the questions buyers would ask, and note the URLs it cites. Those pages are where you need to get in.
2. Let review and ranking authors get real information
Give authors an account they can try, and write price, features and who it is for as one clear page. Do not buy a ranking spot and then pretend it is an independent review.
3. Write your official site so it can be cited directly
Say in the first sentence what you are and who it is for; put price and features in a table; state the updated date.
4. Let crawlers in
Check whether OAI-SearchBot, PerplexityBot and Claude-SearchBot are blocked; see AI visibility audit checklist.
5. Test again with the same set of questions every few weeks
Look only at changes in appearance frequency, not at a single answer.
The full evaluation is in GammaSlide case study.
FAQ
Are there public scoring criteria for AI brand recommendations?
No. The official docs from OpenAI, Google and Perplexity cover crawlers and eligibility for inclusion, and have not published recommendation scoring rules. Everything on this page is labeled as observation or inference.
If I rank high on Google, will AI recommend me?
Not necessarily. In Ahrefs' 2025-08 study, only 12% of cited URLs ranked in Google's top 10 for the original prompt.
Why do third-party rankings matter?
Observation: of the 28 web pages read in our review, 19 were pages listing multiple similar products. Inference: when answering “what products are there”, AI leans toward citing lists written by others.
Can I pay for a ranking spot?
You can run ads, but do not disguise a paid spot as an independent review, and there is no guarantee AI will cite it. The safer way is to let real reviewers get a product they can try and accurate information.
Does AI recommend the same brands every time for the same question?
No. SparkToro's 2026-01 study shows that when the same question is asked repeatedly, the recommended brands are not stable, so you need to sample many times and look at how often you appear.
Sources
- Ahrefs · Only 12% of AI Cited URLs Rank in Google’s Top 10 for the Original Prompt （published 2025-08-11, 15,000 prompts）
- SparkToro × Gumshoe · AIs are highly inconsistent when recommending brands or products （2026-01, 12 prompts run 60–100 times on each platform）
- Pagent · Best AI assistant for your inbox and calendar （The operator’s own site, 2026-11-05）
- Pagent · Methodology （The operator’s own site, 2026-11-05）
- kagentx evaluations: the data on the case study pages; each example gives its date.