What PR Teams Get Wrong About AI Visibility

PR is better positioned for AI visibility than almost any other marketing function, and it is measuring the wrong thing to notice.

The positioning advantage is real: answer engines weight third-party sources heavily, precisely because a brand’s own claims about itself are unverifiable. Earning credible independent coverage is the closest thing to a direct lever on that, and it is what PR already does. Most functions are learning a new job. PR is doing its existing job with a new beneficiary.

The measurement problem is that everything PR reports — placements, reach, share of voice in media, sentiment in coverage — describes what was published, and none of it describes what an engine used. Those turn out to diverge sharply.

Coverage is not citation

A placement is an article existing. A citation is an engine reaching for that article when composing an answer about you. The first does not imply the second, and the reasons are structural rather than about quality.

Some outlets are unreachable. Content behind a hard paywall, or on a site that blocks the search crawlers, may be excellent, prestigious, and entirely invisible to a grounded answer. A tier-one placement your client is thrilled with can contribute nothing to what an assistant says, and nothing in a PR report would reveal that. Should you block AI crawlers covers the publisher side of that decision — and the point here is that your placement inherits whatever choice the outlet made.

Some coverage says nothing quotable. An article that mentions your funding round is a placement. If it contains no claim about what you do, who you serve, or why you would be chosen, there is nothing in it for an engine to use when answering “what is the best tool for X.” Announcement coverage is the largest category of PR output and the least citable.

Recency decays differently. PR treats a placement as a permanent asset. For a grounded answer, retrieval favours what is current; for training-derived knowledge, an article’s influence depends on whether it was in a corpus, which you cannot observe. Neither behaves like the durable credit a coverage report implies.

So the honest framing: coverage is an input with highly variable conversion into citation, and PR currently has no visibility into the conversion rate.

The metric that actually maps

Media share of voice — your mentions versus competitors’ across a monitored outlet set — is the PR metric closest to being useful, and it is measuring the wrong universe.

The universe that matters is not the outlets you monitor. It is the sources engines actually reach for when answering questions in your category, and that set is consistently surprising. It tends to include documentation, community threads, review platforms and comparison sites alongside the trade press, weighted in proportions no media list would predict.

This is knowable rather than theoretical — for grounded answers the engine reports its sources. Citation intelligence is the surface for it, and the first look is usually the most useful hour a PR team spends on this, because it reorders the target list. Pitching the outlet that engines actually cite beats pitching the outlet with the best circulation figure, and those are frequently not the same publication.

What PR is uniquely good at here

Three things, and they are worth stating because the framing above is critical and the function deserves better than that.

Original data placement. The most citable thing a brand can publish is a statistic that exists in exactly one place, and the fastest way to propagate it is coverage. This is the one part of the argument with controlled evidence behind it: the GEO paper found that adding quotations, statistics and cited sources measurably raised a page’s visibility in generative engines, while keyword-oriented edits did not. Journalists need fresh numbers; a genuine survey or proprietary aggregate is the easiest thing to earn coverage for. Each resulting article becomes another source feeding retrieval, under your name, in places you never pitched. This is the highest-leverage play available to a PR team and it requires the data to be real — original research is the highest-ROI AEO play is unambiguous about why a fabricated figure is worse than none.

Correcting the record where it can be corrected. When an engine states something false about a brand, the model cannot be edited. What can change is what retrieval finds and what the third-party consensus says — and moving a consensus across independent sources is precisely PR’s craft. Brand hallucinations are an operations problem covers the triage; the remediation half is largely a communications job.

Consistency of description. Engines resolve entities from how sources describe you. A brand described five different ways across its coverage builds a smeared node; one described consistently builds a solid one. Message discipline across earned media has always been a PR goal and it now has a mechanical payoff.

What PR should stop claiming

Two overclaims that will not survive contact with a sceptical CFO.

“Our coverage drove AI visibility.” Even where presence rises after a campaign, the attribution is not available. Coverage, content changes, competitor activity and model updates all move the same number, and the causal chain from a placement to a mention in a synthesised answer is not traceable. What you can honestly say is narrower and still worth saying: these sources are cited by engines in our category, we earned placements in them, and our presence moved over the same window.

“AI sentiment improved because our messaging landed.” Sentiment in an AI answer is a coarse read on prose that aggregates many sources, including ones you did not influence. It is useful for detecting a change and poor for characterising a cause.

The general principle is the same one that governs the rest of this discipline: presence is observable, influence is inferable, causation is not available. A PR function that reports the first two carefully will be trusted longer than one that asserts the third.

The counter-argument

The strongest objection: PR has spent twenty years being told its metrics are inadequate, this is one more critique from an adjacent discipline, and coverage volume remains the thing clients and executives actually ask for.

That is fair on both counts, and the practical response is additive rather than replacing anything. Keep reporting placements — they are the deliverable and the function is accountable for them. Add one column: whether the outlet is one engines demonstrably cite in your category. That single addition changes targeting decisions without dismantling a reporting structure anyone depends on, and it is defensible because it is measured rather than asserted.

There is also a legitimate case that some coverage is worth earning regardless of citability — recruitment, investor confidence, credibility with partners. Not everything PR does is meant to influence an answer engine, and treating citability as the only lens would be its own error.

Where to start

1. Pull the sources engines actually cite for ten questions your buyers ask. Compare against your target media list.
2. Check reachability on your last ten placements: paywalled, crawler-blocked, or open?
3. Add a citability column to coverage reporting. Do not remove anything.
4. Pitch a number, not an announcement. One genuine data point out-earns three funding stories.

AEO for PR and communications covers where this sits in a comms programme, and PR strategy for AI visibility has the tactical layer.

The one-line version: PR is already doing the highest-leverage work in AI visibility, and measuring it with instruments that cannot see whether it landed.