Here's a scene that's played out in PR engagements all year. A client hands their firm a Google-shaped question (“which publications should we target to rank in ChatGPT?”) and the firm burns a quarter trying to answer something the technology can't hold still long enough to measure.
AI citation sources change 40 to 60% every month. The better question is about co-occurrence: where does your client's brand need to appear next to its category, often enough, that the models learn the association?
I run AI trainings for PR teams, including one inside a global pharma company this June. Smart people. Decades of earned-media instinct. And the question their leadership handed them was the same one you're probably getting from clients right now.
So let's take the question apart, gently, because the people asking it were right for 15 years.
Why can't you rank in ChatGPT?
Because there's no list to rank on.
Google is deterministic. Same query, similar results, stable enough to track weekly and report monthly. AI engines are probability machines. Rand Fishkin's January 2026 SparkToro study with Gumshoe ran 12 prompts 2,961 times across ChatGPT, Claude, and Google's AI. Ask the same best-brands question 100 times and there's less than a 1-in-100 chance you see the same list of brands twice. The same order? Closer to 1 in 1,000.
Fishkin's verdict, verbatim: “any tool that gives a ‘ranking position in AI’ is full of baloney.”
The publication lists churn just as hard. Profound analyzed 240 million ChatGPT citations and found 40 to 60% of cited domains change month to month for identical queries. Over 6 months, 70 to 90% are completely different. And the engines barely agree with each other: FancyAI ran 500 commercial queries across 5 platforms and found only 6.8% of cited domains appeared on 3 or more of them.
One more, because this is the stat that should end the target-publication-list conversation for good. Evertune analyzed 200 million prompts and found that even the single most-cited domain in a category rarely exceeds 5% of total citations. The other 95% spreads across thousands of domains.
Your client's category is a probability cloud, and every engine draws from it differently.
Does a tier-1 placement still get you cited?
Sometimes. Less often than you'd think, and less often than you're charging for.
Muck Rack's “What Is AI Reading?” research, built on 25 million-plus cited links, found a 2% average overlap between the journalists PR teams pitch most and the journalists AI engines actually cite for the same brands. Two percent. Muck Rack's CEO called the gap striking.
Ranking well in Google helps less than it used to, too. Pages ranking #1 in Google get cited by ChatGPT only 43.2% of the time (Kevin Indig with AirOps). Roughly 60% of AI Overview citations come from URLs outside the top 20 organic results.
Here's the twist that should make every PR founder sit up: the same research shows earned media drives 82 to 90% of AI citations. The craft is winning. The targeting and the scoreboard are losing.
What is AEO, and what should you measure instead of rank?
AEO (answer engine optimization) is the work of structuring content and earned presence so AI engines pull it into answers. For a PR firm, the honest, defensible metric underneath it is share of citation: the percentage of AI answers that include your client, measured across a fixed prompt set run many times per platform.
That's the one metric that survived Fishkin's study. Rank is noise. Presence frequency across large samples is real and measurable. And it happens to be a share-of-voice frame, which PR has understood for decades.
On that pharma call, the sharpest question came from their comms lead: “I'm doing an interview with reporters who are going to use AI to write the article. How do I make sure my accurate information survives into the piece?” That is a co-occurrence question. She got there faster than most of the industry.
How do you appear in ChatGPT results for your clients?
Four moves, all inside the work you already know how to do.
1. Rebuild the target list around what AI actually cites. Most AI visibility for a brand comes from about 20 outlets (Muck Rack). They're rarely the legacy tier-1 list. Pull the citations AI gives in your client's category, find the journalists and outlets behind them, and pitch those. Medical journals, review sites, specialist trades, and niche newsletters beat a Vogue mention for citation purposes, and I say that having told a pharma PR team exactly that in June.
2. Chase co-mentions, and lots of them. Brand web mentions correlate 0.664 with AI citation, roughly 3x stronger than backlinks at 0.218 (Ahrefs, 75,000 brands). Heavily cited text averages 20.6% proper nouns against 5 to 8% for normal English (Indig, 1.2 million responses). Named brands next to named categories, across many credible sources. That's the mechanism.
3. Check the plumbing before the pitch. On that same training, the team realized mid-call that their corporate site might be blocking AI crawlers from reading their own press releases. Every placement in the world can't fix a robots.txt file that tells the models to go away. Audit crawlability first. It takes an afternoon.
4. Report share of citation quarterly, and average across runs. A single snapshot is statistically unreliable when 40 to 60% of sources drift monthly. Fixed prompt set, 60 to 100 runs per prompt per platform, presence percentage versus competitors. That's a number a CFO can trust, which matters, because PR budgets get cut when the metric is impressions and defended when the metric is “we appear in 41% of AI answers about the category, up from 12%.”
The part that should make you optimistic
PR firms are sitting on the highest-value input in the AI visibility chain. Earned media drives the overwhelming majority of citations, and only about 16% of brands systematically track AI search performance (McKinsey, September 2025). The window is wide open for the firms that swap the scoreboard now.
Your discipline already works. Point it at a new target list and measure it with a new number.
If you're a founder trying to figure out whether your own brand shows up when buyers ask AI about your category, that's one of the six areas we audit in The Diagnostic. You can start The Diagnostic here, or book The Fit if you want to talk it through first.