Link building for AI visibility: mentions, citations and a plan

AI assistants make a brand visible in two ways. They name it in the answer text, and they link to a page as a source. The first depends on what the model has seen about you across the web. The second depends on what it retrieves when the question is asked. Link building and digital PR affect both, though not in the way a link count would suggest.

One warning before the numbers. Almost all the evidence here is correlational and published by companies that sell AI visibility tools. Treat it as a direction, not a law.

Mentions and citations are two targets

When an assistant searches the web, it rewrites the prompt into several searches, retrieves pages, picks passages and writes an answer with sources attached. That produces two separate outcomes.

• A mention is your brand named in the answer. It comes from the model’s memory plus whatever the retrieved pages say about you.
• A citation is your URL shown as a source. It comes from retrieval.

They overlap less than you would expect. Semrush’s 2026 AI Visibility Index found that on Gemini the overlap between mentioned brands and cited domains was as low as 30%. Track them separately. How AI search chooses sources explains the retrieval steps.

Mentions versus links: what the evidence shows

Ahrefs studied 75,000 brands and published correlations with AI visibility in December 2025.

Factor	ChatGPT	AI Mode	AI Overviews
YouTube mentions	0.737	0.740	0.712
Branded web mentions	0.664	0.709	0.656
Branded anchors	0.511	0.628	0.527
Branded search volume	0.352	0.466	0.392
Domain Rating	0.266	0.285	0.326

In the earlier May 2025 version of the study, on AI Overviews only, referring domains correlated 0.295 and backlinks 0.218.

How to read this honestly:

• How much the web talks about you tracks AI visibility far better than how many links you have.
• This is not causation. Big brands have more mentions, more searches, more links and more AI visibility for the same reason: they are big.
• Links are not irrelevant. Branded anchors, meaning links whose clickable text is your brand name, are among the strongest factors. A link with your name on it is also a mention.
• Links act indirectly as well. Seer Interactive’s January 2025 study of 10,000 questions found backlinks weak or neutral, but ranking on Google’s first page correlated about 0.65 with being mentioned. Rankings depend on links.

The takeaway is “links plus context”. A link in a relevant article that names and describes your brand does two jobs. A link from an unrelated page with a keyword anchor does at most one.

What AI assistants cite most

Peec AI analysed 30 million citations in the US in March 2026. The overall top ten were Reddit, YouTube, LinkedIn, Wikipedia, Forbes, G2, Yelp, Facebook, Medium and TechRadar. Peec states that Reddit and YouTube were the only two in every engine’s top five, although its own published ChatGPT list is Wikipedia, Reddit, Forbes, TechRadar and LinkedIn. Google’s AI surfaces leaned towards social platforms and ChatGPT towards editorial sites.

Three corrections to common claims:

The shares are small. In Profound’s analysis of 680 million citations from August 2024 to June 2025, Wikipedia was 7.8% of ChatGPT citations and Reddit was 6.6% of Perplexity citations. The much larger percentages often quoted are shares within each engine’s top ten sources, not of all citations.

The mix is unstable. Semrush reported that Reddit’s share of ChatGPT citations fell from 3.8% to 0.5% within August 2026, an 86% drop. A strategy built on one platform can lose its footing in a week.

Earned beats paid. Muck Rack’s May 2026 report, covering over 25 million links, put earned media at 84% of citations, journalism at 27% and paid or advertorial content at 0.3%. Muck Rack sells PR software, its “earned” category includes Reddit, Wikipedia and reviews, and we took the figures from a press release, so treat them as reported.

“Best of” lists matter too. Peec AI’s research from September 2025 to March 2026 found that holding the top rank in a cited listicle lifted brand visibility by 16.5 percentage points in B2B software. That is vendor data and correlational.

What counts as manipulation

On 15 May 2026 Google changed its spam policy definition to include “attempting to manipulate generative AI responses in Google Search”. Google described this as a clarification of existing rules. Some coverage also refers to a named policy on “inauthentic mentions”. We did not find a policy by that name on the live page on 3 October 2026.

The practices observed so far, described so you can recognise them:

• Fake community posts. Agencies seeding Reddit threads with recommendations from invented personas. Reddit said its detection caught about 25,000 spam posts and comments a day in the first quarter of 2026.
• Undisclosed paid listicle placements. A market for paid inclusion in “best of” lists is reported at $49 to over $8,000 per placement, a low-confidence range. Disclosed, such content is rarely cited. Undisclosed, it is deception.
• Self-promotional listicles at scale. Brands ranking themselves first on their own blogs. Lily Ray reported in February 2026 that several software sites with many such pages lost 29% to 49% of Google organic visibility from mid-January 2026. She noted it was one factor among several.
• Prompt injection. Microsoft Security reported in February 2026 that it found more than 50 attempts by 31 companies to plant instructions in assistant memory through “Summarize with AI” buttons. Microsoft treats this as a security threat.
• Rented authority. “Best of” pages placed on high-authority hosts, which is parasite SEO under another name and falls under the site reputation abuse policy.

A rule of thumb: if a mention exists only because you paid for it and the reader is not told, assume it is on the wrong side of the policy.

A practical plan

1. Pick your prompts. List the questions a buyer would ask an assistant in your category. A working guide is 50 to 300 prompts per market.
2. Find what gets cited. Run each prompt many times and record the third-party URLs that appear repeatedly. Bing Webmaster Tools shows Copilot grounding queries for free, and the AI visibility tools comparison lists paid trackers.
3. Get onto those pages. Ask to be included, corrected or reviewed on the pages that are cited again and again. Give list editors pricing and comparison data that justifies including you.
4. Fill gaps with original data. Publish research journalists will cite. This is standard digital PR, and the digital PR services comparison lists agencies if you need one.
5. Ask for branded links. When coverage names you, a link with your brand as the anchor is the best outcome. Follow up on unlinked mentions, but do not treat an unlinked one as a failure.
6. Cover video and community. YouTube mentions were the strongest correlate in the Ahrefs data. Get reviewed by creators in your niche, and take part in forums under named, disclosed staff accounts.
7. Keep building relevant links. Rankings still feed retrieval.
8. Measure mention rate, not position. SparkToro’s January 2026 research found the same brand list came back less than 1% of the time across repeated runs. Report how often you appear across many runs, with ranges, and review the trend monthly.

Two things to skip. An llms.txt file showed no relationship with AI citation frequency in SE Ranking’s November 2025 study of 300,000 domains. And Google’s documentation says no special files or markup are needed to appear in its AI features.

Where to go next

Reputation signals beyond links are covered in off-page E-E-A-T and reputation. For any term used here, see the glossary.