AI Visibility for PR Agencies: Shortlist Prompts, Facts and Honest Client Advice

Buyers ask an assistant for a shortlist before they ask an agency for a pitch. The prompts below are hypothetical, but they follow the shape of real agency searches:

• "Which PR agencies in Chicago specialize in B2B SaaS?"
• "Best healthcare PR firms for a Series B medtech launch"
• "Crisis communications agency in London with financial services experience"
• "Compare [Agency A] and [Agency B] for consumer tech PR"
• "Is [your agency] independent or part of a holding group?"
• "Which PR firms in Austin offer media training for founders?"
• "What sectors does [your agency] work in, and where are its offices?"

The first four are discovery prompts. They test whether you get named at all, and they combine a service, a sector and a city. The last three are verification prompts. A buyer runs them after seeing your name, and they test whether the engine describes you correctly.

Your prompt list is a sample, not the real input. OpenAI says ChatGPT search rewrites a prompt into one or more targeted queries before it searches. So write 5 to 8 prompts using your real sectors and cities, then keep them fixed. A stable list is what makes next month's results comparable with this month's.

Marketing leads and founders use several assistants, and each one reaches the web through its own crawler or index. As of October 2026:

• ChatGPT search fetches through OAI-SearchBot. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers.
• Google AI Overviews and AI Mode only use pages that are indexed and eligible to show with a snippet. Google says there are no additional technical requirements and no special optimizations.
• Perplexity indexes through PerplexityBot.
• Claude search relies on Claude-SearchBot.
• Microsoft Copilot grounds its answers in the Bing index.

Search is not always on. When an engine answers without searching, it relies on training data frozen at a cutoff. OpenAI's model docs list a Feb 16, 2026 cutoff for the GPT-5.6 family and Aug 31, 2025 for ChatGPT's Instant path. An agency that rebranded or merged after those dates can show up under its old identity in ungrounded answers.

Engines also diverge sharply. In Otterly's 2026 experiment with two AI-generated blog sites, Microsoft Copilot kept citing them after Google had removed both from its index, while Google's AI features almost never did. Score each engine on its own. The difference between API and app answers matters too, and AI API answers vs consumer app answers explains why a test through an API may not match what a buyer sees.

Start with evidence, not a guess. Run your prompts, then record every URL each engine cites. Those cited pages tell you where the engine is reading about your category.

For PR agencies, cited pages often fall into these groups:

• Trade press agency rankings, profiles and "agency of the year" coverage
• Award shortlists and winners pages
• Agency directories and listing sites
• Client press pages and case studies that credit the agency
• Structured data sources such as Wikidata, plus your LinkedIn company page

Citations concentrate on a small set of sources. Researchers studying AI search (arXiv 2507.05301, data from March to May 2025) found the top 20 most frequent news sources accounted for 67.3% of all citations for OpenAI models. The practical reading: a mention on an outlet engines already cite is worth more of your attention than volume elsewhere. That is an association in the evidence, not a measured lever.

For each cited page, check your name, specialties, offices, leadership and ownership status. Ask the publisher to correct anything wrong. The PR for AI answers playbook covers how to earn mentions on pages engines retrieve.

No published study measures how often engines get agency facts wrong, so treat this as a checklist, not a ranking. Say you run a hypothetical 25-person healthcare PR firm with offices in Boston and London. These are the facts that decide whether a buyer shortlists you:

1. Specialties. An engine that files you under consumer lifestyle drops you from every healthcare prompt.
2. Office locations. A closed office still listed, or a new one missing, breaks city-based prompts.
3. Ownership. Independent versus part of a holding group is a buying criterion, and acquisitions are easy for engines to miss.
4. Current name. A rebrand after a model's cutoff leaves the old name in ungrounded answers.
5. Services. Crisis communications, media training, investor relations and public affairs are separate buying decisions.
6. Clients. A former client described as current is a confidentiality and credibility problem.
7. Leadership. Departed founders often linger.

The fix is plain. State each fact once, specifically, on a page that loads as HTML. Make the about page, contact page, footer and directory listings agree. A specific stated fact gives an engine a sentence it can quote. It does not guarantee the engine uses it.

Access is where the evidence is strongest. If a crawler cannot fetch your site, nothing on it reaches the answer. Check these five things:

• robots.txt, per bot. Rules are independent. GPTBot controls training only. OAI-SearchBot controls ChatGPT search. A blanket AI block often removes search eligibility by accident.
• Firewall and CDN. A firewall can return 403 to a crawler that robots.txt allows. OpenAI says eligibility requires both crawler permission and host or CDN access from its published IP ranges. Cloudflare changed its defaults on 2026-09-15 to block Training and Agent crawlers on ad-displaying pages for new domains and existing free-tier customers.
• JavaScript. Most AI crawlers other than Googlebot do not execute JavaScript. Specialties and office details that only appear after scripts run may be invisible to them.
• noindex and snippet controls. noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers. Templates sometimes apply them by mistake.
• Index status. Check Google Search Console and Bing Webmaster Tools separately.

A minimal robots.txt that allows the main search crawlers looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

To see whether your firewall turns any of them away, the AI Crawler Checker requests a page as each AI crawler and shows the response.

Clients will ask whether coverage gets them into ChatGPT. The honest answer: when engines cite other sites, being named on those sites is associated with appearing in the answer. Nobody has measured a reliable effect. A 2026 critical survey of 45 GEO studies (arXiv 2607.14035) found no technique with a stable, longitudinal, cross-platform causal effect on discoverability.

That honesty is a selling point. Avoid claims the evidence contradicts:

• "Add schema or FAQ schema to get cited." Google says structured data isn't required for generative AI search.
• "Publish an llms.txt." Google says Search ignores it, and no engine documents reading it.
• "GEO delivers up to 40% more visibility." That figure came from a lab engine with content already in context.
• "Your AI ranking." Google exposes AI impressions only, with no position data.

Report what you can show: the fixed prompt list, the date, each engine separately, and whether the answer used search. Note that a citation is not the same as influence, since OpenAI's own docs show engines retrieve more sources than they cite. Agency client reporting for AI visibility covers the template, and clients with no coverage yet can start with how a startup with no press gets mentioned by AI.

Fix your own visibility first. It is the case study clients will ask about.

1. Days 1 to 3: baseline. Write 5 to 8 prompts with your real sectors and cities. Run them logged out in each engine. Record which agencies get named and which URLs get cited. The free checker on this site gives a snapshot for three buyer questions.
2. Days 4 to 7: access. Check robots.txt per bot, firewall responses, JavaScript dependence, snippet directives and index status in both Google and Bing.
3. Days 8 to 14: your own site. Write a one-page fact sheet covering specialties, services, offices, ownership, name and leadership. Make every page on your site agree with it, in plain HTML.
4. Days 15 to 21: third-party pages. Work through the cited URLs from your baseline. Request corrections to directories, award pages and trade profiles.
5. Days 22 to 30: re-run and record. Run the same prompts again. Note changes per engine without claiming cause, since answers vary between runs.

If you resell this work to clients, what agencies should check in white-label AI SEO covers the claims to keep out of the offer.