How AI assistants choose which products to recommend

On this page

1. How does an assistant put an answer together?
2. What tracks with being mentioned by AI?
3. Does ranking in Google get you cited?
4. Which sites do assistants cite most?
5. Does the way a page is written matter?
6. What has no measured effect?
7. How are buyers using assistants?
8. What does this mean for a new SaaS product?
9. Questions
10. Sources

AI assistants recommend products they have seen described across the web, on pages they can read. Across 75,000 brands, Ahrefs found that mentions of a brand on other sites and on YouTube tracked AI visibility far more closely than backlinks did. No study we found shows structured data or an llms.txt file causing a citation.

How does an assistant put an answer together?

An assistant draws on two things: what the model learned in training, and what it fetches while answering.

OpenAI's own crawler documentation shows the split. OpenAI runs three separate user agents. GPTBot collects content for training models, OAI-SearchBot gathers pages for ChatGPT's search results, and ChatGPT-User fetches a page when a person asks for it¹².

Anthropic runs a similar set of three¹³.

Google describes the fetching side in its documentation on AI features. It uses a technique it calls query fan-out, where one question is turned into several related searches and the answer is built from the results¹⁰.

For a new product the two routes behave differently. Training data is slow, and a product launched this month is not in it. Live search is where a new product can appear soon, and that depends on pages that can be found and read today.

What tracks with being mentioned by AI?

Being talked about on other sites tracks most closely with being mentioned by AI. Ahrefs compared 75,000 brands against the things marketers usually measure, and mentions on other sites came out well ahead of links.

A follow-up in December 2025 widened the study to ChatGPT and Google's AI Mode. Mentions on YouTube showed the strongest correlation of all, at about 0.737. Branded web mentions came in between 0.66 and 0.71 depending on the platform².

These numbers are correlations, and Ahrefs says so itself. Large brands score high on every measure at once, so the study can't show that mentions cause visibility.

It is also a vendor study, run on the vendor's own index.

If you have been counting links, count where your product is described as well.

Does ranking in Google get you cited?

Ranking in Google matters less than it used to for Google's own AI answers, and much less for other assistants.

These studies can't be lined up as a trend. Ahrefs changed its method between the two AI Overview studies, so part of the fall from 76% to 38% is a change in how it was measured. BrightEdge counts a page ranking anywhere, where Ahrefs counts the top 10.

Read together, the studies show that Google's AI answers draw heavily on pages Google already ranks. ChatGPT, Gemini, and Copilot mostly cite pages that are not in the top 10 for the question as typed.

Perplexity sat in between, with nearly one in three of its citations in Google's top 10⁶.

For a founder that is encouraging. A new site that can't yet reach Google's first page can still be cited by an assistant.

Which sites do assistants cite most?

Wikipedia, Reddit, and YouTube are the sites AI assistants cite most, and each holds a small share of the total.

Profound analysed 680 million citations between August 2024 and June 2025⁸.

You may have seen much larger figures quoted, such as Wikipedia at 47.9% of ChatGPT citations. That number is real, and it is a share of the ten most cited sources only. Of all citations, Wikipedia's share was 7.8%⁸.

Pew Research Center found something similar for Google. Wikipedia, YouTube, and Reddit were the three most cited sources in Google's AI summaries, and together they made up 15% of the sources listed⁹.

The shares also move. Semrush tracked more than 230,000 prompts weekly for 13 weeks in 2025. ChatGPT cited Reddit in close to 60% of responses in early August and around 10% by mid September¹¹.

For SaaS, G2 appears in the top sources for both ChatGPT and Perplexity, so a review profile is one of the places an assistant may read about you.

With the biggest source below 8%, most citations go to ordinary sites.

Does the way a page is written matter?

One peer reviewed study has tested how page content affects AI answers, and it found that it does, with limits.

Researchers from Princeton and other institutions tested changes to page content against a benchmark of 10,000 queries. Adding citations, quotations, and statistics gave a relative improvement of 30 to 40% on the paper's main visibility measure. Keyword stuffing did not perform well¹⁵.

The test ran mainly on a simulated engine built on GPT-3.5, with data from 2023. Visibility was measured as a page's share of the words in the answer, and clicks and sales were not measured.

The paper's headline figure is "up to 40%", which is a best case.

We treat it as a direction to write in. Give a page numbers with sources, quote people by name, and link to where a fact came from. Nobody should expect a 40% gain from doing so.

What has no measured effect?

Structured data, llms.txt, and rewriting pages for machines are widely recommended, and none has a measured effect on AI citations.

Structured data. Ahrefs ran a controlled test on 1,885 pages that added JSON-LD, against 4,000 control pages. Citations in Google AI Overviews fell by 4.6%. The changes in AI Mode and ChatGPT were too small to tell apart from zero³.

Google's documentation says of its AI features: "There's also no special schema.org structured data that you need to add"¹⁰. Structured data still has a documented role in Google's ordinary search results, so it is worth keeping for that.

llms.txt. Google says the file "will neither harm nor help your site's visibility or rankings in Google Search"¹⁶. Ahrefs looked at 137,210 domains and found that 97% of llms.txt files received no requests in May 2026¹⁷.

Rewriting for machines. Google's guide lists content "chunking" and rewriting for AI among the things site owners can ignore for Google. It lists chasing inauthentic mentions there too¹⁶. That guidance covers Google's products only.

In our own check of 313 new SaaS websites, 53.4% had an llms.txt file and 81.8% had structured data on the home page¹⁸. Most new products have already done the two things with no measured effect on citations.

How are buyers using assistants?

Software buyers use AI assistants heavily for research, and they check what they are told.

G2 surveyed 1,076 B2B software buyers in March 2026. 51% said they start software research with an AI chatbot more often than with Google, up from 29% in April 2025¹⁴. G2 is a review site with an interest in that finding.

TrustRadius surveyed 1,862 technology buyers in January 2026. 63% used AI to research their purchase, and 94% of those fact-checked what it told them¹⁹. TrustRadius is a review site too.

Search has not gone away. SparkToro and Datos found in 2025 that over 20% of Americans were heavy users of AI tools, while 95% still used a traditional search engine each month²⁰.

An assistant may be where a buyer first hears your name, and your own site is where they go to check it. Both have to say the same thing.

What does this mean for a new SaaS product?

A new SaaS product should do five things, and this is the order we would do them in.

1. Make sure your pages can be read. An assistant that fetches a blank page has nothing to quote. Can Google and AI assistants read a site built with AI tools? has the test.
2. Say plainly what the product is. One sentence, near the top of the home page, that names the product, what it does, and who it is for.
3. Get described in other places. Launch directories, a review profile, a walkthrough video, answers in communities where your buyers already ask. Do this honestly. Google lists inauthentic mentions as something to ignore, and buyers fact-check.
4. Publish pages with something of your own in them. Your own numbers, a named customer's words with their permission, a test you ran. This is what the one controlled study on page content supports.
5. Build the pages buyers ask for at the point of choosing. See The comparison and alternatives pages every SaaS needs.

Then keep the facts current, starting with your prices.

Questions

Do backlinks help a product get recommended by AI?

They correlate weakly. In Ahrefs' study of 75,000 brands, backlinks correlated with AI Overview visibility at 0.218, against 0.664 for branded web mentions. These are correlations and do not show cause.

Does schema markup get a page cited by AI?

No study shows that it does. Ahrefs tracked 1,885 pages that added schema and found no major uplift in citations. Google says no special structured data is needed for its AI features.

Does a page need to rank in Google to be cited by ChatGPT?

No. Ahrefs found that 12% of links cited by ChatGPT, Gemini, and Copilot appeared in Google's top 10 for the same prompt.

Which website does ChatGPT cite most?

Wikipedia, at 7.8% of all ChatGPT citations in Profound's analysis of 680 million citations. The often quoted 47.9% is Wikipedia's share of the top ten sources only.

Does llms.txt help with AI visibility?

There is no evidence that it does. Google says the file has no effect on Google Search, and Ahrefs found 97% of llms.txt files received no requests in a month.

How many software buyers use AI assistants for research?

In G2's March 2026 survey of 1,076 B2B software buyers, 51% said they start research with an AI chatbot more often than with Google. In TrustRadius's 2026 survey, 63% of technology buyers used AI during purchase research.

Sources

1. Ahrefs, "An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied)", 26 May 2025. https://ahrefs.com/blog/ai-overview-brand-correlation/
2. Ahrefs, "Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied)", 12 December 2025. https://ahrefs.com/blog/ai-brand-visibility-correlations/
3. Ahrefs, "We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved.", 11 May 2026. https://ahrefs.com/blog/schema-ai-citations/
4. Ahrefs, "76% of AI Overview Citations Pull From the Top 10", 21 July 2025. https://ahrefs.com/blog/search-rankings-ai-citations/
5. Ahrefs, "Update: 38% of AI Overview Citations Pull From The Top 10", 2 March 2026. https://ahrefs.com/blog/ai-overview-citations-top-10/
6. Ahrefs, "Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt", 11 August 2025. https://ahrefs.com/blog/ai-search-overlap/
7. BrightEdge, "AI Overview Citations Now 54% from Organic Rankings". https://www.brightedge.com/resources/weekly-ai-search-insights/rank-overlap-after-16-months-of-aio
8. Profound, "AI Platform Citation Patterns: How ChatGPT, Google AI Overviews, and Perplexity Source Information", 5 June 2025. https://www.tryprofound.com/blog/ai-platform-citation-patterns
9. Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results", 22 July 2025. https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
10. Google Search Central, "AI features and your website", updated 10 December 2025. https://developers.google.com/search/docs/appearance/ai-features
11. Semrush, "The Most-Cited Domains in AI: A 3-Month Study", 10 November 2025. https://www.semrush.com/blog/most-cited-domains-ai/
12. OpenAI, "Overview of OpenAI Crawlers". https://developers.openai.com/api/docs/bots
13. Anthropic, "Does Anthropic crawl data from the web, and how can site owners block the crawler?", 7 April 2026. https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler
14. G2, "New G2 Research: Half of B2B Software Buyers Now Start Their Research With AI Chatbots", 15 April 2026. https://www.prnewswire.com/news-releases/new-g2-research-half-of-b2b-software-buyers-now-start-their-research-with-ai-chatbots-302742807.html
15. Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, and Deshpande, "GEO: Generative Engine Optimization", arXiv 2311.09735, KDD 2024. https://arxiv.org/abs/2311.09735
16. Google Search Central, "Optimizing your website for generative AI features on Google Search", updated 10 July 2026. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
17. Ahrefs, "We Analyzed 137K Sites: 97% of llms.txt Files Never Get Read", 15 June 2026. https://ahrefs.com/blog/llmstxt-study/
18. SaaS GTM, site check of 313 newly launched SaaS websites, 28 September 2026. Own data. Method and totals.
19. TrustRadius, "2026 B2B Buying Disconnect Report" press release, 15 July 2026. https://www.prnewswire.com/news-releases/trustradius-2026-b2b-buying-disconnect-report-reveals-ai-has-changed-how-buyers-research-but-not-what-they-trust-302825792.html
20. SparkToro, "New Research: 20% of Americans use AI tools 10X+/month, but growth is slowing and traditional search hasn't dipped", 26 August 2025. https://sparktoro.com/blog/new-research-20-of-americans-use-ai-tools-10x-month-but-growth-is-slowing-and-traditional-search-hasnt-dipped/