Google AI Overviews: What Changes for PR and Comms Teams

Most of what has been written about AI Overviews is written for SEO teams, and it is mostly about traffic. That is a fair concern and it is not the interesting one.

The interesting one is this: Google now writes a short article about your brand, for free, on demand, at the top of the page, assembled from sources you did not choose. For a communications team that is not a search problem. It is the largest unmanaged piece of published coverage most brands have.

What an Overview actually is

A generated answer that sits above the results for some queries, with links to the sources it drew on. Not every query gets one, availability varies by market, and Google adjusts which queries qualify fairly often.

Two things separate it from a chat assistant, and both work in your favour.

It shows its sources. Almost always, with links. A chat answer may or may not tell you where it came from. An Overview generally does, which hands you the citation list without any work.

It is grounded in the live index. An Overview is not primarily recalling training data, it is summarising pages Google has crawled. That means it moves faster than a model’s memory, and it responds to new publishing on a timescale comms teams can actually work with.

Where the content comes from

This is the part worth being precise about, because there is a persistent misunderstanding attached to it.

Overviews are built from pages Google indexes through ordinary Googlebot. There is no separate crawler to allow, no submission process, no opt-in.

Which leads to the thing teams get wrong in both directions: Google-Extended does not control AI Overviews. It governs whether your content is used to train Gemini. Blocking it will not keep you out of Overviews, and the only way to achieve that is to block Googlebot, which removes you from Google Search altogether. If somebody in your organisation blocked things in a hurry a couple of years ago, this is worth checking before anything else on this page.

The practical entry requirement is therefore conventional discoverability. If a page cannot be found normally, it will not be summarised. Everything else is a matter of whether your page is easy to lift from, which is covered properly in the generative engine optimization pillar.

What actually changes for comms

Earned coverage became a direct input. This is the headline. A trade article about your launch used to do two jobs: reach that publication’s readers, and sit in your coverage report. It now does a third, which is feed the answer Google gives to everyone who asks about your category. A single well-written, well-placed piece can end up being paraphrased to an audience many times larger than the outlet’s own.

That changes how a placement should be valued, and it changes what you ask for in it. A clear, factual, quotable paragraph about what you do is worth more than a flattering adjective, because the paragraph is what gets summarised.

Your worst coverage got more reach too. The same mechanism runs in both directions. An old critical article that still ranks is now a candidate source for an answer about you, which raises the return on getting the record corrected rather than waiting for it to slide down page two.

Correction has a faster clock. Because Overviews read the live index, publishing an accurate, clearly written page can change what appears within weeks. Compared with shifting what a model absorbed in training, that is quick.

Being absent is now a visible outcome. If a category Overview names three companies and you are not among them, you have lost the question before your website was ever a factor.

The measurement problem

Here is where most programmes come unstuck, and it is worth stating plainly rather than working around quietly.

Search Console does not separate AI Overview appearances from ordinary results. Those impressions are folded into your totals, so you cannot isolate how often you appeared in an Overview, or how often you were cited inside one, from the standard reports.

The consequence is a genuine measurement gap, and the wrong reaction to it is to fall back on sessions. On queries where an Overview answers the question outright, publishers widely report fewer clicks, particularly on informational searches. Commercial and navigational queries hold up better.

So a fall in sessions is no longer evidence that your content stopped working. It may be evidence that it started working somewhere you cannot see.

You can become more influential and less visited in the same quarter. Any reporting that treats those as the same trend will tell the board a story that is precisely backwards.

The workable alternative is to measure it directly. Take a fixed set of queries that matter to your category, run them on a schedule, and record four things each time: whether an Overview appeared, whether you were named, whether the claims about you were accurate, and which sources were cited. That is a real measurement with a stated sample, and it is a far more honest slide than a traffic chart.

What to do about it

Audit your crawler rules first. Five minutes, and it is the only item here that can silently undo everything else.

Pick the twenty queries you actually care about. Your category question, your brand question, your main comparisons, the objection questions. Record the Overview for each. This is the baseline, and doing it once is already more than most competitors have.

Read the cited sources. That list is the current editorial board for your brand, chosen by Google rather than by you. Some of them will be outdated, some will be wrong, and a few will be pages you could reasonably get corrected or updated this quarter.

Fix the claims worth fixing at the source. Publish the accurate version on your own site, then get it corroborated somewhere independent. The full sequence is in AI reputation management.

Brief your PR team on quotability. The single highest-leverage change is asking for coverage that states plainly what you do and what is true about you, early in the piece. That paragraph is the one that gets summarised into an answer for people who will never read the article.

Re-run the set monthly. Overviews change as the index changes. One reading is a snapshot of a moving thing.

Who owns this

It is tempting to leave it with search, because it happens on a search results page. That produces a programme optimised for clicks that nobody is reading for accuracy.

The better arrangement is that search owns discoverability and the technical floor, comms owns what the answer actually says about the brand, and somebody owns the monthly re-run so it does not quietly stop in month three. Which is the same conclusion as almost everything else in this cluster, for the same reason: the machinery is technical and the failure is reputational.

Tracking it across engines rather than one query at a time is what AI reputation tracking is built for.

Common questions

How do AI Overviews affect brand visibility?

An Overview answers the question above the blue links, so a reader can learn what Google thinks about your brand without visiting anything. Visibility shifts from ranking to being cited inside the answer, which means you can be more influential and less visited at the same time.

How do you appear in a Google AI Overview?

There is no separate submission or opt-in. Overviews are assembled from pages Google already indexes through ordinary Googlebot, so conventional discoverability is the entry requirement. Beyond that, pages that answer one question directly near the top, in specific and dated language, are easier to lift into an answer than pages that bury the point.

Should I block Google-Extended to stay out of AI Overviews?

It will not work. Google-Extended governs whether your content is used for Gemini model training, not AI Overviews, which read through ordinary Googlebot. The only way to keep a page out of Overviews is to keep it out of Google, which almost nobody actually wants.

Can you see how often you appear in AI Overviews?

Not cleanly in Search Console, which does not break Overview appearances out as their own dimension, so those impressions sit inside your ordinary totals. Practical measurement means running a fixed set of queries on a schedule and recording whether an Overview appeared, what it said and which sources it cited.

Do AI Overviews reduce traffic?

Publishers widely report fewer clicks on queries where an Overview answers the question outright, particularly informational ones. Commercial and navigational queries are affected less. The important consequence for comms is that a drop in sessions is no longer reliable evidence that your content stopped working.