A journalist quotes you in a trade publication. Six weeks later, someone asks an AI assistant a question in your category and your name comes back in the answer — sourced from that article, not from your website. You did not optimise for that. You earned it.
The takeaway up front: AI answer engines are, in large part, a re-ranking of the same earned media you were already chasing — so the work that gets you cited in AI answers is mostly PR work, not a new technical discipline. The publications that AI systems lean on are the publications journalists write for. What changes is not the tactic; it is which kind of coverage compounds, and how you measure it, since none of your existing PR reports will show you an answer nobody clicked.
This guide covers what actually gets pulled into an AI answer, how to make your organisation quotable at the sentence level, and a measurement routine you can run monthly without buying anything.
What changed, and what did not
Assistants like Google's AI Overviews and AI Mode, ChatGPT's search mode, Perplexity, and Microsoft Copilot answer a question in prose and then attribute that prose to a handful of sources. The user often reads the answer and never clicks. That is the part that unsettles PR teams — coverage can now generate influence that produces no referral traffic at all.
What has not changed is the supply side. These systems are summarising the open web, and the open web's most quotable material is journalism, trade press, expert roundups, and reference pages. A brand that appears in credible third-party writing is a brand that appears in the summary. A brand that only appears in its own marketing copy usually is not.
Two practical consequences follow:
- Third-party mentions matter more than owned pages, because a summary needs an attributable source and your own site is the weakest possible one.
- The unit of value moves from the article to the sentence. An AI system does not cite "that great feature in a trade magazine." It reproduces a specific claim. Whether your claim is the one that survives compression is now part of the craft.
The coverage types that get pulled in
Not all earned media is equally extractable. In rough order of usefulness for this purpose:
- Roundups, comparisons, and "best of" lists in independent publications. These map directly onto the commercial questions people ask assistants ("what's a good tool for X"). Being in the list body — with a described strength — beats being a logo.
- Trade and vertical press. Narrower publications are disproportionately useful for narrow questions, and they are far easier to earn than national coverage. This is the same logic behind digital PR and high-authority links: relevance often outperforms raw reach.
- Expert quotes in explainer articles. When you are quoted defining or explaining something, your sentence is pre-formatted for extraction. This is the highest-leverage return on the least effort, which is why answering reporter source requests is worth systematising.
- Original data you published and someone else reported. A number that only exists because you produced it becomes the canonical answer to a question, and every outlet that reports it reinforces the association.
- Reference and community pages — industry wikis, association directories, well-moderated forums. Unglamorous, durable, frequently cited.
What underperforms: undifferentiated press-release syndication, brand-voice thought leadership with no verifiable claim in it, and paid placements that read like paid placements. None of these give a summariser anything to hold on to.
Make your organisation quotable at the sentence level
If extraction happens sentence by sentence, then write sentences that survive being lifted out of context.
Give every announcement one extractable claim. Not an adjective — a fact. "Our platform is the leading solution for mid-market logistics" cannot be quoted by anything. "We process shipments for 400 mid-market logistics operators across 12 countries" can. If you cannot state the claim without a superlative, you do not have a claim yet.
Put the definition in the quote. When a reporter asks what something means, answer in a complete, self-contained sentence that would still make sense with no surrounding paragraph. Assume it will be pasted somewhere with your name attached and nothing else.
Standardise your descriptor. Decide the one clause that describes what you do, and make sure every spokesperson, boilerplate, and pitch uses it identically. Inconsistent self-description is the most common reason a brand shows up in answers with the wrong category attached to it.
Publish the underlying data somewhere stable. If your number is worth reporting, it needs a permanent page with the methodology on it, so the outlets that cover it link to a source that stays put.
Keep your basic facts current everywhere. Founding details, headcount range, headquarters, product category, leadership names. Assistants assemble these from whatever the web says, and stale press-kit copy propagates for years.
A measurement routine that survives scrutiny
You cannot report on this the way you report on clips, and pretending otherwise is how PR teams lose credibility with a CFO. The honest approach is a small, repeatable panel — the same discipline described in the PR measurement guide, applied to a surface with no impression data.
Build a prompt panel. Write 15 to 30 questions a real buyer would ask: category questions ("best options for…"), comparison questions ("X versus Y"), problem questions ("how do I fix…"), and brand questions ("is [your brand] any good"). Freeze the wording. The panel only works if it does not drift.
Run it on a fixed cadence, in a clean session. Monthly is enough. Use a logged-out or fresh session so personalisation and memory do not flatter you. Run each prompt across the two or three assistants your audience actually uses.
Record four things per prompt: whether you were mentioned, whether you were cited with a linked source, which source was cited, and which competitors appeared. Which source got cited is the most actionable column on the sheet, because it tells you exactly which article to go earn more of.
Report it as share of voice, not as a count. "We appeared in 9 of 30 category prompts, up from 5, and competitor B appeared in 21" is a defensible sentence. "AI mentions up 80%" from a base of five is not.
Cross-check with referral data. In your analytics, watch for referrals from assistant domains — chatgpt.com, perplexity.ai, copilot.microsoft.com and similar. The volume is usually small compared to search, but the sessions tend to arrive with unusually high intent, which is worth noting separately rather than burying in "other." Google's AI surfaces are harder to isolate, since Search Console does not give them a separate report, so treat the prompt panel as your primary instrument and referrals as supporting evidence.
Do not convert it into a money figure. The temptation to price these mentions is strong and the arithmetic is fiction. The same caution applies here as in calculating earned media value: report exposure and movement, model influence separately, and never present an estimate as revenue.
What not to do
Do not buy your way into listicles on low-quality sites in the hope of being scraped. It works briefly, it is visible to anyone who checks your backlink profile, and it attaches your brand to sources you would be embarrassed to see cited.
Do not publish statistics you cannot substantiate. A fabricated number that gets picked up becomes permanent and is trivially checkable, and the correction costs more than the coverage was worth.
Do not build a parallel "AI SEO" programme with its own budget and its own agency. The inputs are the same inputs: relevance, credibility, real news, and a spokesperson who says specific things. Fund the PR programme and change what you measure.
FAQ
Is this the same as SEO? It overlaps, but the lever is different. Technical work helps your own pages be readable; it does not put you in an answer that cites someone else's article. The dominant input is third-party coverage, which makes it a PR function with a measurement problem attached.
How long does it take to show up? Longer than a campaign cycle. Sources have to be published, indexed, and then favoured by a system you do not control. Judge it over a quarter with a fixed prompt panel, not week to week.
Should I add structured data or an "AI policy" page? Clean, accurate structured data on your own site is good hygiene and worth doing. It is not the mechanism that gets you into a summary of what other people wrote about you, so do not let it displace outreach work.
What if the AI describes my company wrongly? Trace the source. Wrong descriptions almost always come from a stale article, an outdated boilerplate, or an inconsistent category claim. Fix the facts at the source — updated press kit, corrected article, consistent descriptor — and re-run your panel next cycle.
Do press releases help? A release that carries a genuine, checkable claim and gets picked up by real outlets helps, because the pickup is the citable artefact. Syndication alone, with no editorial pickup, rarely does anything here.
The brands that show up in AI answers are the ones journalists were already quoting. Get the coverage first, make the sentences worth lifting, and measure it honestly. Find the right journalists and open press opportunities on PR Rush, and turn the coverage you earn into the sources those answers quote.
