Key takeaways
- AI engines learn about your brand from the whole web, not just your site. Independent corroboration is what turns a claim into a mention.
- News coverage is a short-lived citation asset. In Promptwatch Data, half of the news pages ChatGPT cited on July 6, 2026 had dropped out within 36 days, so pair PR bursts with durable pages.
- Product pages (22.4%) and landing pages (16.8%) take the biggest share of ChatGPT citations, ahead of listicles (13.6%). PR should push people toward those pages and into the roundups AI already cites.
- Write extractable, evidence-backed claims in the first 100 words of every placement. Vague narrative does not get absorbed into answers.
- Measure citations against a locked prompt list, per engine, every week. Hit counts tell you nothing about AI visibility.
Why PR changed when AI search arrived
Most PR reports still count placements. Reach, impressions, a logo wall. AI engines do not care about any of that. They care whether what was written about you is specific enough to quote, and whether several independent sources say the same thing.
The raw material is already earned media. Muck Rack's May 2026 study of more than 25 million links cited by AI found that 84% were earned media, with journalism alone at about 27%. So PR is feeding the machine whether you plan for it or not. The question is whether your campaigns are built to be picked up.
A recent paper on arXiv splits AI citation behavior into two stages, and I find this split useful for PR planning:
- Citation selection: the engine retrieves your content and decides to cite it.
- Citation absorption: the engine uses your language, numbers, or structure inside the generated answer.
Most campaigns only think about selection. A TechCrunch piece that describes you in fuzzy language gets selected and never absorbed. A data-dense mention in a mid-tier trade publication can do more. That is a slightly uncomfortable truth for anyone who has spent years chasing tier-one logos.

Step 1: Baseline what AI says about you today
Do not brief a single journalist before you know where you stand. Write down 20 to 50 questions a buyer in your category actually asks. Include category questions ("best payroll software for small teams"), comparison questions, and a few branded ones. Run them in ChatGPT with search on, Perplexity, and Google AI Mode, and record which domains each answer cites and whether you are named.
Lock that list. Changing prompts week to week makes every trend meaningless. Answers vary run to run, so a single screenshot of a good answer proves nothing.
Doing this by hand works for the first week. After that, a platform is easier. Promptwatch tracks prompts across the real interfaces of ChatGPT, Perplexity, Google AI Overviews and others, and its citation reports show which third-party pages AI leans on in your category, which is exactly the target list a PR team needs.

If you want a free first look, the free AI visibility report scans how ChatGPT, Gemini and Perplexity describe your brand.
Step 2: Map the sources AI already trusts in your category
PR for AI visibility is a targeting problem. Pull the cited domains from your baseline run and sort them into families: news, trade press, review sites, listicle publishers, community platforms, reference sites.
Then compare against what Promptwatch Data says about engine behavior, because the families differ a lot by engine:
- In Promptwatch's social citations report (Oct 11 to Nov 9, 2026), social platforms made up 13.16% of Google AI Overviews citations, 1.94% of Perplexity's and 0.64% of ChatGPT's. YouTube alone was 6.71% of AI Overviews citations.
- For ChatGPT, the same report points to listicles, guides and comparisons rather than social posts.
- Promptwatch's Reddit data shows reddit.com's share of ChatGPT citations falling from roughly 4% to 0.5% on August 14, 2026. A community-seeding plan built for last year's ChatGPT is stale.
The practical upshot: a Reddit thread may help you in AI Overviews and barely register in ChatGPT. Choose channels per engine, not per habit.
Step 3: Aim PR at the content types that get cited
Promptwatch Data on which content types ChatGPT cites (Sep 28 to Oct 4, 2026) is the most useful PR planning input I have seen this year.
| Content type | Share of ChatGPT citations |
|---|---|
| Product pages | 22.4% |
| Landing pages | 16.8% |
| Listicles | 13.6% |
| How-to guides | 8.7% |
| Documentation | 7.5% |
| General articles | 6.1% |
| Comparisons | 5.9% |
| News | 5.7% |
| FAQ pages | 3.3% |
| Reviews | 1.7% |
Two things stand out. News is only 5.7%. And on branded prompts, press releases still earn 2.1% of citations, higher than their share across all prompts, so they have a small but real job when someone asks about you by name.
For commercial prompts like "best CRM for small businesses", product pages (26.0%), landing pages (20.6%) and listicles (15.0%) dominate, and comparisons rise to 7.9%. The report's advice is blunt: get into the roundups ChatGPT cites, collect reviews on trusted sites, and publish comparison pages.
For PR, that translates into three jobs. Land placements in the listicles and roundups your baseline run surfaced. Make sure your own landing and product pages hold the facts that coverage will reference. And give journalists a reason to link to those pages.
Step 4: Plan for the half-life of coverage
Here is where I would change most PR calendars. Promptwatch's half-life study tracked every page ChatGPT cited on July 6, 2026 for 90 days.
- News: almost a quarter of pages were never cited again after day 0, half were gone after 36 days, and 41% were still cited at day 60.
- Landing pages and docs: 78% and 76% still cited after 30 days, 69% and 67% after 60.
- Comparisons: 72% after 30 days, 63% after 60.
- Product pages, reviews and how-tos: about 55% still cited after 60 days, reaching half-life at 74 to 77 days.
- Listicles: 60% still cited after 60 days.
The report's own conclusion is that news and press coverage suits short bursts of visibility. I would go further: treat a launch story as the spark and the durable page as the fuel. Every announcement should point to a landing page, comparison or doc that stays cited for months.
Step 5: Engineer corroboration across source families
Vendors repeat a claim that brands mentioned positively on four or more unaffiliated platforms are 2.8x more likely to appear in ChatGPT answers. The origin is a Prompt Insider post that does not publish its method, so treat it as a direction rather than a threshold. The underlying logic is sound anyway: an engine trusts a claim more when different kinds of sources agree.
So spread a campaign across families rather than blasting one wire. A reasonable mix for a single launch:
- One national or business news story
- One or two trade publication pieces with the actual data
- A roundup or listicle inclusion
- A review or directory profile that states the same facts
- Optionally, a YouTube explainer or podcast guest spot, which matters most for AI Overviews
Keep the core facts identical everywhere: what you do, who it is for, the headline number. Inconsistent descriptions are how engines end up repeating the wrong thing about you.
Step 6: Write placements that can be absorbed
Getting selected is not enough. The pitch and the press release need to hand reporters claims that survive being lifted out of context.
- Put a specific, evidence-backed claim in the first 100 words. Numbers, dates, named methodology.
- One idea per paragraph. If a paragraph only makes sense after the previous one, an engine cannot quote it.
- Name the entity plainly: company, product, category, location. Skip taglines.
- Publish original data. A proprietary dataset gives journalists something they cannot get elsewhere and gives AI something unique to cite.
- Link the placement back to a canonical page on your site that states the same facts in plain text.
Semrush makes a similar point in its digital PR for AI visibility guide: your own claims look self-serving until trusted third parties repeat them.
Step 7: Fix your own site so citations have somewhere to land
Coverage that points to a thin or blocked site wastes the effort. Check that AI crawlers can actually read your pages. The Agentic Readiness Check shows which crawlers can access your site, and Promptwatch's AI crawler logs show whether bots such as GPTBot or PerplexityBot fetch the pages your PR points at, and where they fail.
Treat this as a prerequisite. A great mention in a respected outlet does little if the page it links to returns errors for AI bots.
Step 8: Measure citations, not clips
Replace the clip report with a citation scorecard. Review it weekly against your locked prompt list.
| Metric | What it tells you | How to track |
|---|---|---|
| Citation presence | Are engines citing campaign content? | Weekly prompt runs per engine |
| Mention rate | Are you named in answers at all? | Visibility score or share of voice |
| Source mix | Which families drive your citations? | Citation reports by domain and type |
| Sentiment and claims | What does AI say about you? | Brand narrative analysis |
| AI referral traffic | Do citations become visits? | Analytics, UTM and referrer segments |
Track each engine separately. If citations fall across many prompts in one week with no site change, the engine probably changed its retrieval, not your page. In GA4 you can watch referral sessions from chatgpt.com and perplexity.ai, though note that Promptwatch's visitor analytics covers attribution in more depth now that ChatGPT stopped sending a referrer on October 8, 2026.
Tools worth a look
You can do this with a spreadsheet at first. As volume grows, these are the options I would compare. For a wider list, the GEO software directory at bestgeosoftware.com covers more.
| Tool | Best for | Notes |
|---|---|---|
| Promptwatch | Tracking, tracing sources, then acting | Citations, crawler logs, visitor analytics, content agents |
| Semrush | Teams already in the Semrush suite | Cited pages view and AI visibility reports |
| Ahrefs Brand Radar | Broad citation benchmarks | Monthly view of most-cited domains |
| Meltwater | Classic media monitoring | Coverage tracking, not AI citations |
| Otterly.AI | Basic, affordable prompt tracking | Monitoring only |


Meltwater and similar tools are still useful for knowing who wrote about you. They just do not tell you whether an AI engine used it.
Step 9: Close the loop with durable content
When the baseline shows a gap, say a roundup that never mentions you, or a comparison page you do not have, fill it. Promptwatch's Content Agents and content gap analysis are built for this: they find the prompts where competitors are cited and you are not, then draft pages to cover them. Whatever tool you use, the point stands. Coverage buys the burst, owned pages keep the citation.
A 90-day plan
- Weeks 1 to 2: lock the prompt list, take the baseline, audit crawler access.
- Weeks 3 to 4: map cited sources by family and engine, pick 20 target publications and roundups.
- Weeks 5 to 8: run one data-led campaign across four source families, with a canonical landing page behind it.
- Weeks 9 to 10: publish comparison and FAQ pages for the gaps the baseline exposed.
- Weeks 11 to 12: re-run the baseline, compare per engine, and decide what to repeat.
For a fuller structure, the 90-day GEO program in the Promptwatch Academy goes deeper on sequencing.
Common mistakes
- Counting placements and calling it success.
- Sending the same release to 50 outlets on one wire.
- Describing the company differently in every pitch.
- Putting the evidence in paragraph six.
- Relying on news coverage alone when it decays in about five weeks.
- Using ChatGPT behavior to guess what Perplexity or AI Overviews will do.
- Trusting vendor statistics without a method behind them. The 2.8x and 33% overlap figures floating around come from vendors, not independent studies.
Hiring help
If you would rather hand this to specialists, look for an agency that shows you the locked prompt list, the start date, and per-engine before and after results. 1001 SEO Media combines digital PR with GEO work for this reason, and runs its AI visibility tracking on Promptwatch. Whoever you pick, walk away from anyone who guarantees citations.
The short version
Pick targets from the sources AI already cites. Give reporters claims that can be quoted. Spread coverage across source families. Back every burst with a durable page. Then measure weekly against the same prompts. A clip that no engine ever repeats is a nice screenshot and nothing else.
