Does AI cite my own website or third-party sites more often?

It depends on how 'own' is defined and which questions you ask. In the studies we verified, the share going to brand-owned sites runs from a few percent, when only the brand's own page counts, to more than half, when any brand's site counts. Third parties tend to lead questions that compare or recommend brands, while factual questions about your brand favour your site. Measure your own split per engine and per query type.

Why does ChatGPT cite Reddit and G2 instead of my website?

For evaluative questions, such as which tool fits a team or how two products compare, an answer is easier to support with independent opinions than with a vendor's own claims. Review platforms and forums gather many such opinions in one place. Your site is more likely to be cited for facts only you can state, such as pricing, specifications and policies. Check that the cited threads and reviews describe you accurately.

Should I keep publishing on my own site if AI mostly cites third parties?

Yes. Your site is the expected source for pricing, specifications, policies and documentation, and it is where you can publish the correct version of a fact that a third party will not fix. Build it alongside external presence: reviews, independent coverage and accurate community discussion. Decide how to split effort by measuring which query types each one wins for your brand, engine by engine.

How do I check whether AI engines are citing my own site?

Write down what counts as your site, then build a fixed set of brand questions and evaluative questions. Run each prompt several times on each engine, copy every cited URL from the sources the answer shows, and classify each one as your site, a competitor or a third party. Report the split per engine and per query type, date the baseline, and repeat on a schedule.

Can AI tools generate citations for my own writing, and are they reliable?

That is a different meaning of the word citation. AI assistants can draft references for essays or reports, but a generated reference may point to a source that does not exist or does not say what is claimed. Open every reference, confirm the document exists and check that it supports the sentence before you use it. This article covers AI engines linking to web pages instead.

What Share of AI Citations Come From Your Own Site, and Why It Varies

In the studies we verified, the share of AI citations going to brand-owned sites runs from a few percent to more than half. Three choices explain the range: how 'own' is defined, which engines are measured, and whether questions concern the brand or choosing between brands. Below are the sourced figures and a method to measure your split.

What counts as an own-site citation, and why definitions decide the number

This article is about AI engines such as ChatGPT, Perplexity, Gemini and Google's AI features linking to web pages in their answers, not about AI tools generating references for academic writing.

Own site, competitor sites and third parties: three buckets, not two

A first-party, or own-site, citation is a linked source on a domain your brand controls. A third-party citation is a linked source on a domain someone else controls. That two-way split hides a group that matters in every comparison question: competitors. A competitor's product page is third party from your side, yet it is owned content for another brand. Count three buckets instead: your site, competitor sites, and independent third parties such as publishers, review platforms and forums.

Write down the choices that move the result before you count:

Brand social profiles. They sit on someone else's domain but carry your content. Decide whether they count as own.
Marketplace listings. A product page on a retailer or app marketplace is written by you and hosted by them.
Documentation subdomains. A docs or help-centre subdomain is usually own, but state it, because some tools match only the main domain.
Partner pages. Reseller and integration pages often reuse your copy. Classify them one way and keep it fixed.
The unit. Counting citations, unique URLs, or answers with at least one own-site link gives different percentages from the same data.

A share published without these choices is hard to compare with anything.

Citation versus mention

A citation is a linked source attached to the answer. A mention is your brand named in the answer text. An answer can name you without linking to you, and it can link to your page while recommending someone else. This article counts citations. For the full distinction, see how an AI mention differs from an AI citation.

What share of AI citations go to brands' own sites, study by study

We found two studies whose own pages state an own-site or brand-site share, both consulted on 10 November 2026. Each figure is shown with its definition of 'own'.

Study and publisher	Date	Engines	Sample	Definition of 'own'	Own-site share	Third-party share
Where do AI citations come from?, Profound	Page dated 30 July 2026	All models in its dataset; per-engine list Not verified	Not verified	Any brand site	Roughly 57% brand citations	Not verified
Duplicate Content and AI Citations, Shero Commerce	Page dated 13 July 2026	Not verified	1,851 sources	A brand's own page	2.8%	59% publishers; other third parties Not verified

Profound reports that "globally across all models, AI citations are roughly 57% brand citations". Shero Commerce reports that "across all 1,851 sources, just 2.8% were a brand's own page while 59% were publishers". Publishers take most citations in Shero's dataset; brand sites take more than half in Profound's.

Neither study has to be wrong. Profound's wording counts any brand site, so on our reading a competitor's site also counts as a brand citation, while Shero counts only the brand's own page. The query mix adds to the gap: when a prompt asks which product to buy, the answer can cite several brands and publishers, so one brand's own page is a small slice even when brand sites together are large. Both points are our inference from the quoted wording, not statements on either page.

We also looked for published community-platform and earned-media shares; none could be verified on the publisher's own page as worded, so none is used here.

To compare two benchmarks, or your data with a benchmark, align the unit (citations, URLs or answers), the buckets (own, competitor, third party), the engines, and the mix of brand and evaluative questions. If one of these differs, say so next to the number. For the wider picture, see which domain categories AI engines cite most.

Why the split changes with the engine, the question and the brand

The engine and how many sources it cites per answer

Engines differ in which sources they cite and in how many they attach to each answer. We did not find a per-engine count we could verify on a primary page, so treat it as something to measure. The logic holds as an inference: an engine that attaches many sources leaves room for your page beside publishers and competitors, while an engine that attaches few makes your page compete for scarce slots. That is how one engine can show a short list of cited pages for a question and another a long one.

A page must also be retrieved before it can be cited. Google Search Central states, in documentation consulted on 10 November 2026: "To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements." Eligibility is not selection, and other engines run their own retrieval. A page that ranks on Google can still be missing from what ChatGPT or Perplexity retrieves for a prompt, so Google ranking alone does not predict an own-site citation.

Questions about your brand versus questions about choosing a brand

Factual questions about a brand, such as its price, integrations or refund policy, have an obvious primary source: the brand's site. Evaluative questions, such as which tool suits a team or how two products compare, ask for a judgement, and engines tend to draw it from independent sources. A prompt set weighted toward brand questions should report a high own-site share, and one weighted toward category questions a low one. This is reasoning from how the questions differ, not a measured ratio.

Known brands versus niche brands

As an inference, a well-known brand tends to have enough independent coverage for third parties to fill its evaluative answers, while a niche brand may appear only through its own site, or not at all, in category questions. We found no sourced own-site split for local businesses, for non-English markets, or for ecommerce broken down by engine. Treat those splits as open questions to measure on your own prompts, not as figures to estimate.

Why third parties win evaluative answers, and where your own site still wins

What the vendor documents. The Google passage quoted above sets eligibility for supporting links; it does not describe how they are chosen.

What a study measured. Ahrefs, on a study page dated 26 May 2025, reports that "Web mentions (0.664) correlate much more strongly than backlinks (0.218)" for brand visibility in AI Overviews. That is a correlation, not proof of cause, and it covers one Google feature.

What we infer. When a question asks for a judgement, a claim repeated across independent sources is easier for an engine to support than a claim found only on the vendor's site. Corroboration is the likely reason third parties lead evaluative answers. Review platforms such as G2, and Reddit and other forums, are likely third-party sources because they gather many independent opinions in one place.

When a cited Reddit thread mentions your brand, check what it says against the facts. If it is wrong or outdated and the community rules allow brand replies, answer from a clearly identified brand account with the correct fact and a link to the page that states it. Do not post through unlabelled employee accounts or invented customers: astroturfing breaks trust and tends to make the thread worse.

When an AI answer cites a wrong third-party page and the publisher will not correct it, escalate in order. Publish the correct fact plainly on an owned page, in a sentence an engine can quote. Correct your entity data where you are described, such as business profiles, directory listings and structured data. Ask other independent pages that cover the topic to state the fact; see our method for outreach to the pages AI engines already cite. Then re-measure the same prompts.

Your own site remains the expected source where only you can state the fact: pricing, specifications, policies and documentation. An engine answering what a plan costs should cite you, and it can only do that if the page exists, is current and can be retrieved. Build those pages alongside external presence, not instead of it.

How to measure your own, competitor and third-party citation split

AI answers are generated afresh for each request, so the cited sources can change between runs of the same prompt and from week to week. A single check is a sample, not a measurement: monitor the split over time.

In ChatGPT, Perplexity and Gemini, open the sources or citations shown with a web-backed answer and copy every linked URL. In Google AI Overviews and AI Mode, expand the supporting links. Interfaces change, so record the date and the interface used with every run.

Use one standard and keep it fixed:

Prompt set. A fixed list split into brand questions and evaluative questions, written the way your buyers ask.
Runs. Several runs per prompt per engine, with the same count every cycle.
Classification. Every cited URL goes into own, competitor or third party, using your written definition of 'own'.
Reporting. The split per engine and per query type. A blended number hides the pattern: a strong own-site share on brand questions can mask near absence on evaluative ones.

To test whether a new third-party placement, such as a review or an article, changed the split, design the test before the placement goes live. Record a dated baseline. Keep a control set of prompts the placement should not affect. Use the same run count before and after. Write the decision rule in advance, for example: the placement counts as working only if citations to it or to your site rise on the target prompts while the control prompts stay flat. Without the control set, normal variation between runs can look like an effect.

Where to start this week

Write your definition of 'own': main domain, subdomains, social profiles, marketplace listings, partner pages, and the unit you count.
Build a prompt set of brand questions and evaluative questions; a couple of dozen is enough to start.
Run each prompt several times on ChatGPT, Perplexity, Gemini and Google's AI features, and copy every cited URL.
Classify each URL as own, competitor or third party, and compute the split per engine and per query type.
Pick one evaluative prompt where a competitor or a wrong third-party page is cited, and choose one action: an owned page that states the missing fact, or outreach to a page that already gets cited.
Record the dated baseline and set the re-test date, with the same prompts and run count.

To see which sites ChatGPT, Perplexity, Gemini and Claude cite for your brand's questions, run a free Namedrop scan as your first baseline.