What is the token launch PR that gets into AI answers?
Write the token launch release as a plain answer to the buyer's question, then place it with independent outlets, so that an engine's own search retrieves it. A Chen et al.1 preprint found retrieval rank explained 32% of which same-site pages ChatGPT cited.
Which sources do AI engines cite most when buyers ask about this?
New Map's measurement2 found reddit.com cited in 20 of 146 answers (13.7%). The questions were 68 about AI visibility tools, asked on 3 engines. That reading held no token-launch questions, so it shows how engines source a discovery question, not a token launch.
On the same date, New Map recorded youtube.com in 16 of 146 answers (11%). Perplexity cited a source in only 7 of its 68 answers, and Claude Code's 10 answers are too few to read alone. How New Map measures sets out the method.
The chart below comes from New Map's measurement and ranks the domains cited in the most of those 146 answers, so it answers which sources lead on discovery questions. Read it as one week's pattern, not a trend.
Does retrieval or page writing decide whether a launch release is cited?
Retrieval decides first. A Chen et al. arXiv preprint found that where a page came back in ChatGPT's own search explained 32% of which pages from one site it cited. All page features combined explained 2.8%.
ChatGPT cited only pages its own search had retrieved, on sub-queries the model rewrote for each answer. Two page traits predicted citation: title match and paragraph match with the question. Page speed, summary boxes, structured data and FAQ blocks showed no effect within a site.
Our read: a launch release is cited when its title and opening paragraphs answer the rewritten question, not when it carries extra markup.
The preprint also reports that the engine's own results on rewritten sub-queries rarely coincide with Google's or Bing's. AirOps3's undated playbook disagrees on structure, reporting citation rates of 2 to 2.8 times for pages with clean headings, FAQs and schema over unstructured ones.
The chart sets the two explained shares side by side. It answers the question: retrieval rank counted for far more than page features.
Why do third-party sources matter more than the release itself?
AirOps's undated playbook puts the third-party share of brand mentions in AI search at 85%. Kevin Indig4's post, which relays AirOps, gives the same 85% for brand visibility in LLMs.
Muck Rack's May 2026 study of over 25 million cited links across ChatGPT, Claude and Gemini put journalism at 27% of citations on unbranded questions. Muck Rack warns that its "earned" label covers every non-owned, non-paid source, so the figure does not show that PR drives those citations.
Ahrefs (December 2025, 75k brands) found YouTube mentions correlated about 0.74 with AI visibility, branded web mentions 0.66 to 0.71 and Domain Rating 0.27 to 0.33. Brand strength confounds that correlation.
Otterly5's analysis of 5.85 million citations found the most recommended brands drew heavily on pages that compare, rank and review their market.
Count sources by origin. New Map's review of crypto AI-visibility research (October 2026) treats ten syndicated copies of one release as one claim, because aggregators inherit their upstream's facts.
How do you write and place a token launch PR for AI answers?
Work through these steps in order. Each one ties to a finding above.
- Make the title and first paragraph a plain answer to the question a buyer would ask about the token. Title and paragraph match predicted citation in the Chen et al. preprint.
- State an explicit price and a current date. Vishwakarma et al. (Sprinklr, SIGIR 2026, 252,000 trials, six models, synthetic pages) found topic match, an explicit price, a current date and being listed first decided which of two retrieved pages was cited first.
- Fix wrong facts at their origin, not at each aggregator that copied them.
- Place the story with independent outlets and comparison pages, the kind of sources behind the most recommended brands in Otterly's 5.85 million citations.
- Refresh the page. AirOps's playbook (undated) found pages refreshed in the last quarter were cited at three times the rate of older ones.
- Re-read the answers every week on each engine and log which sources appear. Plans and pricing show how New Map runs that weekly reading.
The diagram shows how a launch release moves from page to answer.
What limits what a launch PR can do?
Answers move from run to run. AirOps's undated playbook measured brand visibility holding from one run to the next for 30% of brands, and across five consecutive runs of the same prompt for 20%.
Engines also differ. Kevin Indig reports that 91% of citations showed up on just one of ChatGPT, Perplexity or AI Overviews. A source visible on one engine may be absent on the others, so plan sources per engine.
Seeding has a policy edge. Anthropic's usage policy update prohibits using Claude to seed content that misstates its origin, authorship or independence, such as networks of fake-independent sites. iPullRank6's reading says it covers political and commercial deception alike, and the policy takes effect on 12 November 2026. Our read: seed only outlets you can genuinely earn.
Semrush7 (October 2026) notes that the generative AI report in Google Search Console shows impressions only, with no prompt-to-click attribution. Reading the answers directly fills that gap, so the weekly reading in step 6 is how you see what a launch changed.
What to do
- Rewrite the title and first paragraph of your token launch release so each answers one buyer question about the token directly, with the price and a current date stated.
- List the independent outlets and comparison pages that could carry your launch story, and pitch each a distinct angle rather than syndicating one release across all of them.
- Ask the token launch question on ChatGPT, Gemini and Perplexity every week and log the cited sources. Treat moves under about 5 points as possible chance (Peec AI, September 2026). Compare with which sources AI engines cite most.
Frequently asked
Does a press release itself get cited by AI engines?
Rarely: press releases were 1.1% of cited links (Muck Rack, May 2026). Coverage by independent outlets is the better route to a citation, so pitch journalists and comparison pages a distinct angle instead of relying on the release alone.
Should I add FAQ blocks and schema to a launch page?
The sources disagree. The Chen et al. preprint found FAQ blocks and structured data showed no within-site effect, while AirOps's undated playbook reports citation rates of 2 to 2.8 times for structured pages. Treat markup as hygiene, not the plan.
How often does an AI answer repeat itself?
Not often. AirOps's undated playbook found 30% of brands stay visible from one run to the next, and 20% across five consecutive runs of the same prompt. Read answers over many runs, not one.
Sources
- Chen et al. (arXiv), What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO), arxiv.org, 10 October 2026
- New Map measurement, How we measure, 19 October 2026
- AirOps, The GEO Strategy Playbook for Marketers and Organic Growth Teams, airops.com
- Kevin Indig, How to plan 2027 when traffic is down and AI won't attribute, growth-memo.com, 5 October 2026
- Otterly, OtterlyAI and Getfluence Partner to Turn AI Search Citations Into Published Content, otterly.ai, 8 October 2026
- iPullRank (Mike King), Claude Got Guidelines, Now What?, ipullrank.com, 12 October 2026
- Semrush, How to build an AI visibility dashboard with Semrush, semrush.com, 12 October 2026
- Chen and Koudas (University of Toronto), Location, Location, Generation: Where You Ask Matters for Generative Engine Optimization, arxiv.org, 16 October 2026
See where you stand
Book a call and find out how visible your brand is across the AI engines your customers already ask.
Book a callFor agencies: See the agency package