Four AirOps numbers worth keeping

AirOps, a measurement firm working in this space, put out a run of studies recently. Four numbers stuck with me. Each comes from its own piece of work, so treat them as four separate readings rather than one stacked case. Pointed the same way, though, they describe the job precisely: AI visibility is jumpy, it goes stale, it is mostly won on other people’s pages, and communities matter more than most brands realise. Here they are, and here is what we actually do with each.

One: visibility is jumpy

Only around 30% of brands stay visible from one AI answer to the next, and just 20% show up across five runs of the same question. Read that twice. Ask an AI engine the same thing five times and four in five brands wobble in and out of the answer.

This is the single most important fact about measuring AI, and the one most tools quietly skip. A single snapshot lies. If you check once, see your name, and relax, you have measured nothing reliable. It is exactly why an AVS run is not one reading. We ask upwards of 6,000 prompts over a seven-day window precisely because one reading bounces around. It is the only way we have found to see past the wobble and give a client a number they can actually trust. Our tech partner’s own data points the same way: a large share of brands appear only once across a week on unbranded prompts. Appearing once is not the same as being part of the answer set.

Two: stale pages fall out

Pages that have not been updated in over a year are roughly twice as likely to lose their AI citations. Content that sits untouched quietly drops out of the answers, and you rarely notice until a competitor has taken your spot.

So freshness is not optional. We tend to push clients toward a quarterly refresh of the content that matters. But the answer is not a firehose of cheap automated content, because that will not cut through and the engines are getting better at spotting it. It matters because what works is real writing, sometimes started by AI and then finessed by good copywriters, kept moving so you do not slip out of the answers without ever knowing why. The cost of doing nothing is invisible right up until it is not.

Three: other people’s pages do the work

You are around 6.5 times more likely to be cited on someone else’s page than on your own. Your website is not where most of the battle is won.

This is the finding that reorders most marketing budgets, because most marketing budgets are still spent almost entirely on the brand’s own site. The site still matters as the place the answer lands once someone is convinced. But the convincing happens elsewhere: in trade press, in reviews and listicles, in analyst notes, in the third-party sources the engines lean on. It is why we score source quality as a full third of the AVS picture, and why good PR people who know how to earn you a mention in the right third-party places are worth their weight. Getting you named where AI is already looking is exactly the sort of work we can help with.

Four: communities count, but not everywhere

Community sites make up nearly half of all AI citations, and Reddit alone turns up in over a fifth of answers. Communities clearly matter, and we do include guidance on them.

But this is the number most likely to be misused. “Half of citations come from communities” becomes “we need a Reddit campaign” in about ten seconds, and that is usually the wrong move. It is an aggregate across every category. In your particular category Reddit might not show up at all. Across recent work with 25 to 30 different businesses we have seen categories where community sites drive the answers and categories where they are nowhere. So check before you pour effort in. And whatever you do, do not astroturf. Communities spot it, the bans follow you, and the brand ends up as a cautionary tale. What works is monitoring, genuine participation from people who actually belong there, and content good enough that people share it because they want to.

Four readings, one job

A quick caveat worth repeating: these are four separate studies, not one dataset, and AirOps is a commercial firm selling into this market. Take the numbers, question the framing, and do not treat a vendor’s customer stories as independent benchmarks. We read the four as directional, not gospel.

But pointed together they describe the job almost exactly as we built it. Measure repeatedly, because one snapshot wobbles. Keep content fresh, because stale pages drop out. Earn third-party coverage, because that is where citations live. Understand your own category’s sources before spending a penny on communities. None of that is a dashboard you can buy off a shelf. It is a programme, run by people who know what to look for.

Be Known. Be Cited.