Do AI engines trust review sites and directories? Where third-party profiles matter

How much do AI engines cite third-party sites?

Less than most listings vendors suggest, and more than most brands plan for. In Profound's study of 11.84 billion citations across eight AI engines (16 April to 16 July 2026), about 57% went to company-owned websites.¹ The remainder went to third parties: publishers, communities, directories, review sites and reference works. Profound sells AI visibility tracking.

The split depends on the engine. In the same data, about 69% of Gemini's citations went to company sites, against about 47% of ChatGPT's.¹ So a brand whose buyers mostly use ChatGPT relies more on what others publish about it.

It also depends on the industry, and Profound counts competitors' and adjacent companies' sites as company-owned too, so your own site's share is smaller than the headline. Our guide to setting a citation target per engine covers the engine and industry split in depth. This post looks at the third-party share: which review and directory profiles are worth your time.

Share of AI citations going to company-owned sites

Gemini¹69%

All eight engines¹57%

ChatGPT¹47%

Which third-party sites do the engines lean on?

Communities and reference sites lead, with professional networks close behind. Semrush's weekly snapshots from 14 July to 12 October 2025 found Reddit and Wikipedia were ChatGPT's two most-cited domains.² In the same data, LinkedIn appeared in about 15% of Google AI Mode responses.² Semrush sells search and AI visibility tools.

Review sites and directories sit lower in those league tables, but they matter in a narrower set of answers. When a prompt asks for "the best", "alternatives to" or "top-rated" options, an engine needs independent evidence that a vendor is good. Ratings, review counts and category listings supply it. A fair comparison page of your own can compete for the same prompts, as our guide to X vs Y comparison pages sets out.

Our view: treat review sites as evidence for comparison answers, not as a general visibility lever. A complete profile on the review site your category uses does more than fifty thin directory listings. Our notes on Wikipedia and LinkedIn cover the other two kinds of third-party source.

Why buyers' habits make profiles count

B2B buyers now start research in AI and act on what it says, so the evidence an engine finds shapes the shortlist. In G2's survey of 1,076 B2B decision makers in North America, EMEA and APAC (March 2026), 51% said they start software research in an AI chatbot more often than in Google.³ In the same survey, 69% said an AI chatbot had led them to select a different software vendor than they initially planned.⁴

G2 runs a software review site, so it has an interest in this finding; weigh it accordingly. The direction is still useful. If an engine builds its comparison from review profiles and yours is empty or out of date, the answer reflects a competitor's evidence, not yours.

Which profiles are worth keeping complete?

Keep complete the profiles that the engines already cite for your prompts, and the ones your buyers use when they shortlist. Everything else is optional. The way to find out is to run your buyer prompts and record every third-party domain in the answers and citations.

For most B2B companies the short list is the same four kinds: one or two review sites for your category, the main industry directory or association listing, your LinkedIn company page and your Google Business Profile if you serve customers locally. For consumer brands, marketplaces and the review platforms shoppers use take the directory's place.

Where to spend profile effort

Keep complete

• Review sites cited for your category's prompts
• The main industry directory or association
• LinkedIn company page
• Google Business Profile, if you serve locally

Leave alone

• Paid directory bundles with no citations
• Listings on sites no engine cites for you
• Duplicate profiles you cannot maintain
• Review swaps or incentives that breach platform rules

How to find the third-party share for your category

Measure it on your own prompts, because category averages hide the sources that matter to you. Take the questions buyers ask before they shortlist, run them in each engine your buyers use, and record every domain that appears in an answer or a citation. Sort the domains into your own site, competitors' sites, review sites, directories, communities and publishers.

Run each prompt more than once. Answers vary between runs and between days, so a single pass can mislead you about which sources an engine trusts. In a Sigzen AI audit we run each prompt on three separate days across six engines: ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. We use official APIs where they are offered and a licensed AI-answer data provider otherwise.

The output is a short table: each third-party domain, how often it appears, for which prompts and in which engines. That table, not a vendor's list of "top directories", decides where profile work goes. Our synthesis of citation studies by industry is a useful starting hypothesis to test against it.

What does a complete profile look like?

A complete profile matches your website on the facts an engine compares: name, category, what you sell, who you serve, where you operate, pricing model and founding date. Inconsistency is the common failure. If a directory still lists a discontinued product or an old price, an engine may repeat it, as our guide to fixing what ChatGPT gets wrong describes.

Reviews need a steady, honest flow rather than a burst. Ask customers at a natural moment, such as after onboarding or a renewal, and reply to critical reviews in plain language. Never buy or gate reviews; platforms remove them, and engines that read the profile see the pattern.

Our view: a quarterly profile check takes an afternoon and prevents most third-party errors. List each profile, compare it with your site, fix the gaps and note the date. Entity consistency across directories and profiles is part of the AEO/GEO sprint on our AI visibility page.