GEO link building: which sources AI answers cite, study by study

GEO link building is link building done with AI answers in mind: getting a brand named and linked on the third-party pages that systems such as Google's AI Overviews, ChatGPT and Perplexity draw on. Do links matter for generative engine optimization? In the dated studies below, yes: the measures most closely associated with a brand appearing in AI answers are mentions of the brand on other sites, anchors that carry the brand name, and a place on the kinds of page those answers cite, and a link placement can supply each of those.

This page sets out where the term comes from, what eight dated studies measured, which pages AI answers cite, how much traffic those answers send, what Google, OpenAI and Perplexity document, how GEO differs from traditional link building, three routes with numbers, and what we check on a placement. Every figure below is a correlation or a count from one sample, not a cause, and each keeps its own sample, measure, AI system and date.

Where the term GEO comes from

GEO, short for generative engine optimization, comes from the paper that named GEO, by Pranjal Aggarwal and five co-authors, first submitted to arXiv on 16 November 2023, last revised on 28 June 2024 and published in the proceedings of the KDD conference in August 2024. A generative engine, in the paper's sense, is a search system that writes its answer in prose and names the sources it used, where a classic results page returns a list of links to open. The paper framed GEO as a way for content creators to improve how visible their content is in those answers.

The authors built a benchmark of 10,000 queries, called GEO-bench, and tested changes to a page's content. Their central result: "including citations, quotations from relevant sources, and statistics can significantly boost source visibility, with an increase of over 40% across various queries". Two things about that figure matter for this page. It was measured on the paper's own benchmark engine, with a further test on Perplexity that showed improvements of up to 37%, and it tested changes to content, not links. In other words, the founding paper was about what a page says, not who links to it. The studies of links and mentions came later, from software companies and research groups working with live AI systems, and they are what the rest of this page is about.

What the studies measured about links, mentions and AI answers

Eight dated studies have measured links, mentions and the pages AI answers cite. Each row of the table gives one study's sample, what the study measured and what it found, and each figure belongs to that row alone.

Eight studies of links, mentions and AI answers, each figure from the sample named in its row

How to read the table. The first three rows are Spearman correlations, and Ahrefs says in each study that correlation is not causation. A Spearman correlation of 0 means no association and 1 means a perfect one; it shows two measures moving together, not that one causes the other. Brand web mentions are appearances of the brand's name on other sites, branded anchors are those mentions used as the clickable text of a link, and brand search volume is how often people search for the brand by name.

Our reading follows. The two measures at the top of the first row, mentions of the brand on other sites and anchors that carry the brand name, can both come from a placement on a relevant publication, and a branded anchor is a link. The link measures of the brand's own site in the same row, Domain Rating, referring domains and backlinks, sat between 0.218 and 0.326. The third row adds where the mention sits: in AI Overviews, mentions on pages with more referring domains went with visibility at 0.70.

The fourth row measures referring domains against ChatGPT citations, from a model of likelihood. In SE Ranking's study of ChatGPT citations: "Sites with over 32K referring domains are 3.5x more likely to be cited by ChatGPT than those with up to 200 referring domains." The figure covers ChatGPT only, and it comes from a model across 129,000 domains, not a promise about any one site; in that model, the number of referring domains came top of its list of factors. For background on the two measures at the top of the table, see what brand mentions are and the benefits of building backlinks.

Which pages AI answers cite

AI answers cite third-party pages far more than brand-owned ones. Third-party here means a page the brand does not own: a news article, an encyclopedia entry, a social thread, a list on someone else's site. Chen and colleagues at the University of Toronto, in a preprint dated 10 September 2025, found that AI search engines cited third-party sources far more than brand-owned pages, a sharp contrast with Google's more even mix. In their US automotive example, a web-enabled GPT model drew 81.9% of its sources from what the study calls earned media, meaning independent media, review and comparison sites, and 18.1% from brand-owned pages, against 45.1% earned in Google's top ten for the same queries. Muck Rack's May 2026 report, built on more than 25 million links from ChatGPT, Claude and Gemini responses, put about 84% of citations in its earned media class, which it defines broadly as journalism, academic research, government and NGO sources, encyclopedic sites, social platforms and third-party corporate content; journalism alone accounted for about 27%, so the class is far wider than press coverage.

Lists are a large share of the sources. Glen Allsopp's study for Ahrefs, published 4 December 2025, looked at 750 ChatGPT prompts and 26,283 source URLs, and best-of lists on blogs were 43.8% of all page types among the sources ChatGPT drew on. A list is also the kind of page a brand can join by arrangement, which is why lists return in the routes below.

Citation sets also move. Ahrefs measured 37.9% of URLs cited in AI Overviews also appearing in the first 10 results blocks in March 2026, against about 76% in its July 2025 analysis. A single month's picture is a snapshot.

How much traffic AI answers send

In Google's results, searches with an AI summary drew fewer clicks on result links than searches without one, and visits from AI platforms are growing from a small base. Pew Research Center studied 68,879 Google searches by 900 US adults in March 2025, published on 22 July 2025: "Users who encountered an AI summary clicked on a traditional search result link in 8% of all visits. Those who did not encounter an AI summary clicked on a search result nearly twice as often (15% of visits)." Two later measures put numbers on referral traffic. SE Ranking, on 18 June 2026 across 101,574 websites, put AI platforms at 0.32% of all website traffic, up from 0.24% in 2025 and 0.02% in 2024. Similarweb, on 3 September 2026, counted an average of 770.7 million referral visits a month from AI platforms between June 2025 and May 2026, more than double the year before. Both figures measure visits, not citations: a brand can be named in an answer and see no visit from it, so being named in the answer is the main outcome for now.

What Google, OpenAI and Perplexity document

Google's Search Central page "AI features and your website", last updated 10 December 2025, states: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." The same page says both features may use a query fan-out technique, issuing several related searches across subtopics to build a response. One question can therefore send the system looking at many angles of a topic at once, and the pages drawn on can cover those subtopics, not only the exact words of the question.

OpenAI and Perplexity each document the crawler behind their search results. OpenAI's documentation says OAI-SearchBot is the crawler used to surface websites in ChatGPT's search features, while GPTBot crawls content that may be used to train its models. Perplexity's documentation says PerplexityBot surfaces and links websites in its search results and is not used for training. Both pages are undated. Both pages recommend allowing the search crawler in robots.txt for a site to appear in search results, and OpenAI's page says each setting is independent, so a site can allow OAI-SearchBot and refuse GPTBot. For the mechanics behind crawling and indexing, see how a search engine works.

GEO vs traditional link building

The placement itself does not change. GEO link building is still a link in the body copy of a relevant page that people read, under an anchor you chose, on a site with its own readers. What changes is the brief: three habits from the studies above sit on top of the same work.

First, use the brand name in or beside the anchor: branded anchors correlated at 0.527 with brand mentions in AI Overviews in the Ahrefs study of 26 May 2025, and the anchor is something the buyer controls when ordering a placement. Second, look at whether the host page is the kind that gets cited, such as a list or a third-party article on the question buyers ask. Third, count mentions as well as links when you report: a link with a branded anchor counts as both. The three Ahrefs correlation studies counted AI mentions in Ahrefs Brand Radar, and XL PR, below, ran the same unbranded questions through the same engines before and after. None of this ranks one route above another. It is the same work, pointed at the pages AI answers draw on, with the brand name carried where the anchor can carry it.

Link building for GEO, route by route

Three routes map onto what the studies measured: listicle placements, guest posts and curated links, and digital PR. They are set out in no order of merit, each is tied to the rows of the table it speaks to, and the right mix depends on what a brand already has in place.

Listicle placements

Peec AI analysed nearly 200,000 AI responses across eight engines, collected between September 2025 and March 2026 and published on 14 May 2026. The wording of Peec AI's listicle study for its own sample runs: "Visibility: Being featured in a cited third-party listicle heavily influences whether your brand is mentioned in an AI-generated answer." First place in a cited list went with 16.5 percentage points more visibility in B2B SaaS and 13.4 in emerging MarTech. That is Peec's wording and Peec's sample across three industries, not a general law.

Rhino Rank Listicle placements put a brand on a best-of or comparison list. The product is the same on every tier, and the tiers differ only by the minimum Domain Authority of the site.

Guest posts and curated links

Guest Posts are articles of 750 words or more written by our team and placed on an independently owned site in your niche, with one contextual link. Curated Links add a link inside a page that is already published. The difference between the two is whether the page is new or existing. In both, you supply the target URL and the anchor text, which is where a branded anchor comes from.

Tie these routes back to the table. Referring domains, the SE Ranking measure, and branded anchors, the Ahrefs measure, are the two measures these routes add to: a guest post or a curated link on a relevant site can add a referring domain and puts a link under an anchor you chose, which can carry the brand name.

Digital PR

Two single-client accounts come from agencies that sell the work. Kelsey Libert, cofounder of Fractl, a content marketing and digital PR agency, describes a 12-month programme for a matchmaking service that paired answer-first content with digital PR, tracked in Ahrefs Brand Radar across AI Overviews, AI Mode and ChatGPT. The case study is undated: "It earned 1,223 AI citations and 295,561 monthly AI answer impressions, roughly five times the nearest competitor."

XL PR, a PR agency, reported on 8 September 2026 that after four coordinated inputs, including commissioned research, a survey of 1,000 US workers, a wire press release and a trade press feature, its client appeared in 8.8% of 68 answers, built from seventeen unbranded questions across four engines. The agency adds that it does not know how many placements it takes to change an AI answer. Both accounts pair original material, research or answer-first content, with coverage on third-party outlets, and each account is one client, reported by the firm that sold the work, so neither reads as a typical result. For how this route compares with ordered placements, see digital PR compared with curated links.

What we check on a placement

We check the site before anything is placed, and we check the page after it goes live. Before placement, we vet a site on relevance, referring domains, Domain Authority, organic traffic and its trend, and indexed pages, weighed together, with human sign-off before anything goes out. We also check the HTTP status code your target URL returns, so no order points at a dead page or a redirect chain. After publication, we check that the page is live, that the link sits in the body copy, that the anchor and target URL match the order, and that the page is indexable. Only then is the URL reported to the dashboard. Each delivered link carries 12-month replacement cover, so a link that is removed or a page that is taken down is replaced like for like. The full standard is on our quality assurance page.

What we do not claim matters as much as what we check. No study we read has measured whether an ordered placement is followed by an AI citation, and we do not promise one. What a placement does give you is what the studies count: a link under an anchor of your choice, which can carry the brand name, on a relevant third-party page.

Frequently asked questions

What is GEO link building?

GEO link building, or generative engine optimization link building, is link building aimed at the third-party pages AI answers draw on: getting a brand named and linked on the sites that systems such as Google's AI Overviews, ChatGPT and Perplexity cite. The term GEO comes from the 2023 arXiv paper by Pranjal Aggarwal and five co-authors.

Do links matter for generative engine optimization?

In the dated studies, yes. Ahrefs' study of 75,000 brands with Domain Rating over 40, published 26 May 2025, found branded anchors correlated at 0.527 with brand mentions in AI Overviews, and SE Ranking's study of 24 November 2025 found sites with over 32,000 referring domains 3.5 times more likely to be cited by ChatGPT than those with up to 200. Both are associations, not causes.

How is GEO different from traditional link building?

The placement is the same: a link in the body copy of a relevant page under an anchor you chose. GEO adds to the brief: the brand name in or beside the anchor, host pages of the kind AI answers cite, such as lists, and counting mentions alongside links when you report.

Which pages do AI answers cite most?

Third-party pages. Muck Rack's May 2026 report on more than 25 million links from AI responses put about 84% of citations in its broad earned media class, and Chen and colleagues at the University of Toronto, on 10 September 2025, found a web-enabled GPT model drew 81.9% of its sources from independent media, review and comparison sites in their US automotive example. Lists feature heavily too: best-of lists were 43.8% of page types among ChatGPT sources in Glen Allsopp's Ahrefs study of 4 December 2025.

Does Google have special requirements for AI Overviews?

No. Google's Search Central page on AI features, last updated 10 December 2025, states there are no additional requirements and no special optimisations needed to appear in AI Overviews or AI Mode.