Reputation guide

Digital PR for AI: earning a place in the sources AI cites

Digital PR for AI means earning real, accurate mentions on the sites assistants already use as sources in your category: comparisons, media, directories, reviews and forums. It isn’t about buying links or manufacturing mentions, which Google treats as spam, but about giving those sources reasons to talk about you. GeoRush shows you which sites they are and helps you reach them.

Updated on

Folded newspapers, a cup of coffee and a phone on a rustic wooden table

Why third-party sources carry so much weight

When an assistant with search recommends options, it rarely relies only on each brand’s own website. It reads comparisons, “best of” lists, reviews, forums and press coverage, and builds its answer from what those sources repeat. If five comparisons in your market name three competitors and not you, the assistant is unlikely to include you, however good your site is.

That is why GEO (generative engine optimization) has a PR component: identify the sources AI uses for your questions and earn a place in them with accurate information. It is classic digital PR with a different target list: not the most authoritative sites in the abstract, but the ones AI actually cites when your customers ask.

Google adds an important nuance in its generative AI optimization guide: seeking inauthentic “mentions” across the web isn’t as helpful as it might seem, because its systems prioritize quality content and block spam. Its spam policies also explicitly include attempts to manipulate generative AI responses in Search.

Step 1: find the sources AI cites

  1. 1

    Use your buying questions

    Start from the 20 to 50 questions your customers ask. Ask them in several assistants with search (ChatGPT, Perplexity, Gemini, Google AI Overviews), several times.

  2. 2

    Log every cited domain

    For each answer, note the sources: domain, exact URL, and whether they name you.

  3. 3

    Classify by type

    Comparison, directory, reviews, forum, video, media, niche blog, documentation. Each type calls for a different tactic.

  4. 4

    Prioritize

    The best opportunities are sources that show up in many answers, across several questions and engines, and that don’t mention you today.

Two women shaking hands across a desk in a bright office

Step 2: one tactic per source type

Source typeWhat usually worksWhat to avoid
Comparisons and “best X” listsWrite to the author with up-to-date facts, offer a demo or trial, flag factual errors about you with evidence.Paying to be included without it being labeled as sponsored.
Directories and review platformsComplete and verify your profile; ask real customers for reviews systematically.Buying reviews, writing fake ones, or incentivizing only positive ones.
Forums and communitiesTake part with an account clearly identified as the brand, answering questions helpfully.Anonymous accounts recommending you, repeated promotional posts.
Trade media and pressOriginal data, studies, available expert spokespeople, real news.Press releases with no news; paid articles with links that pass ranking credit.
Video (reviews, tutorials)Offer your product to creators who cover your category, with editorial freedom.Undisclosed paid collaborations.
WikipediaIf you meet its notability criteria, suggest changes on the talk page while disclosing your conflict of interest.Editing your own company’s article directly, or paying someone to do it undisclosed.

Step 3: give them a reason to cover you

An author or journalist doesn’t include you because you ask, but because it helps them do their job better. What works best:

  • Original data: a study, a survey or aggregated figures from your operations that nobody else has, with methodology.
  • Accurate, current information: up-to-date prices, plans and features, on a page the author can link to.
  • Access: a demo, trial account or sample so the author can test it themselves.
  • Available experts: people on your team who answer with judgment and speed.
  • Useful resources: templates, calculators or guides the author wants to link because they serve their readers.
  • Honesty: admitting who you are not the best fit for makes everything else more credible.

Email template for the author of a comparison

Short, specific and useful to the recipient. Always personalize it: a generic email shows.

comparison-email.txt
Subject: Updated details for your [category] comparison

Hi [name],

I read your comparison “[title]” ([URL]). It’s one of the most cited pages
when people ask about [category], so I wanted to offer something useful.

We’re [brand], [what you do in one sentence]. If it fits a future
update, here are the key facts:
- Price: [price and date]
- Who it’s for: [customer type]
- How we differ: [1-2 verifiable differences]
- Where we’re not the best fit: [honest]

Happy to set you up with a trial so you can see it yourself.

Thanks,
[name, role, link]

What is allowed and what isn’t

Some “AI PR” practices sold today are spam under Google’s policies or outright illegal in some countries. Always check the rules that apply in your market:

PracticeStatusReference
Sponsored content or paid linksAllowed if labeled; paid links should carry rel="sponsored" or nofollow.Google: qualify outbound links
Buying or selling links for rankingSpam under Google’s policies.Google: spam policies
Paid articles with links that pass ranking creditSpam under Google’s policies.Google: spam policies
Manipulating Google’s generative AI responsesSpam under Google’s policies.Google: spam policies
Fake or purchased reviewsBanned by the FTC in the US since October 2024; may also breach consumer law elsewhere.FTC: rule on reviews and testimonials
Editing Wikipedia about your companyWikipedia asks you not to edit in your own interest; paid editing must be disclosed.Wikipedia: conflict of interest

How to measure whether PR is working

The end goal isn’t “getting a link” but getting AI to mention you more often and more accurately. Measure before and after each action: mention rate on the questions where that source appeared, the position you are named in, and whether the source starts being cited alongside your brand. Be patient: an updated comparison has to be recrawled before it influences answers, and the effect on training-based answers takes longer.

In GeoRush, the get cited feature organizes this work: it lists the sites AI cites without naming you, with an opportunity score, finds public contacts, drafts a pitch email with Claude (3 credits) and runs a pipeline (to contact, contacted, replied, won). When AI starts citing you from that site, the opportunity is marked as won automatically. The sources and citations view shows which source types weigh most in your category.

Common mistakes

  • Chasing high-traffic media that AI doesn’t cite for your questions.
  • Sending the same generic email to a hundred sites.
  • Asking only for “a link” instead of offering useful information.
  • Ignoring errors about you in existing sources: fixing a comparison that is already cited is worth more than appearing in a new one nobody reads.
  • Not measuring: without before-and-after data you can’t tell which actions work.

Which sites does AI cite when people ask about your category?

GeoRush shows you the sources each engine uses, and which of them don’t mention you yet.

Sources

Official documentation and primary sources checked on November 17, 2026. These policies change often: check the source before making important decisions.

Frequently asked questions

It is earning real, accurate mentions on the sites AI assistants use as sources in your category. It differs from classic PR in that targets are chosen by measuring which sources AI actually cites.

Google treats buying links for ranking, and attempts to manipulate its AI responses, as spam, and says inauthentic mentions aren’t as helpful as they seem. Sponsored content is legitimate when it is labeled as such.

Ask your buying questions in several assistants with search, log the domains they cite, and prioritize the ones that appear often and don’t mention you. GeoRush builds that list automatically.

Wikipedia asks people not to edit in their own interest and requires paid editing to be disclosed. The right approach is to disclose your conflict of interest and suggest changes on the talk page.

It depends on when the page is recrawled and whether the assistant searches the web. On engines with search it can show within weeks; for training-based answers, months. These are estimates: measure before and after.

Keep reading