Most of the AI Overviews UK SEO advice doing the rounds stops at the explainer: here is what an AI Overview is, here is a screenshot, now book a call. That does not help you decide what to publish next quarter. The useful question is narrower: which posts in your plan are now being answered for free above the blue links, which ones gained value because a model cannot compress them, and how you tell the difference in your own data rather than someone else's sample.

What those AI Overview percentages actually measure

They are not Google statistics. Google does not publish how often AI Overviews appear. The percentages you see quoted come from third-party SERP trackers (Semrush Sensor, Advanced Web Ranking and similar tools) that run a fixed keyword set through search results on a schedule and count how often a feature fires. Change the keyword set, the device, the country or the month and the number moves. Treat any single figure as an estimate with a sample behind it, not a measurement of your market.

The more useful finding across those trackers is consistent even when the headline number is not: AI Overviews cluster on informational and question-shaped queries, and appear far less on transactional, local and branded searches. So the impact on your blog depends almost entirely on your keyword mix. A Bristol accountancy firm ranking for "r&d tax credit claim accountant"may see Overviews on a small slice of its money keywords while its glossary posts get flattened. Same site, two completely different stories.

UK results also diverge from US samples. Rollout timing differs, local packs and shopping units occupy different amounts of space, and the regulatory backdrop is its own variable: the CMA designated Google with strategic market status in general search in October 2025, and its ongoing case work on general search and search advertising may shape what UK results look like over the next few years. We wrote about one strand of that in our piece on CMA choice screens and what SMEs should do about them. The practical implication for your plan is simple: do not re-plan a quarter off a US-sampled chart.

How to work out your own exposure in about an hour

Open Search Console, set the date range to the last three months, filter country to United Kingdom, and export your top 50 queries by impressions. Then search each one in a clean browser session on google.co.uk, logged out, and record three things in a spreadsheet: does an Overview fire, which domains does it cite, and is the query informational or transactional. Overviews fire inconsistently for the same query, so repeat the ambiguous ones a day later. You now have a real exposure map instead of a vibe. If your rank tracking tool records SERP features by location, switch it to a UK data centre and let it log Overview presence alongside position; most tools track it now, but plenty of accounts are still set to US defaults from the day they were configured.

The blog topics that lost value, and the ones that gained it

Definition posts are the casualty. If the honest answer to your target query fits in three sentences, the Overview will give those three sentences away and the reader has no reason to scroll. "What is a content management system", "what is churn rate", "what does CPC mean": these were never strong commercial assets, and the summary layer has removed what little top-of-funnel value they had.

What still earns the click has a common property: the answer cannot be compressed without losing the thing the reader came for.

• Pricing with real UK figures. Numbers, ranges, what changes the quote, what a quote excludes. Readers click to check the detail against their own situation.

• Comparisons that need a judgement call. Not a feature grid, a recommendation with reasoning and a stated set of criteria.

• Processes tied to a specific tool, version or CMS. Exact menu paths, named settings, what breaks and how to fix it.

• Anything requiring first-hand experience. What went wrong, what it cost, what you would do differently.

Keyword-gap work changes shape as a result. The old instinct was to chase every gap a competitor ranks for. Add a filter before anything reaches the calendar: is this query answerable in a paragraph? If yes, it belongs in a glossary entry, not a 1,400-word post. Our guide to finding content gaps through competitor keyword tracking covers the mechanics of the export; the filter is the new part.

A 12-post quarter, re-weighted

Say your quarter was five definition posts, four how-tos and three list posts. Here is the version we would run instead: cut four of the definition posts from the calendar, repurposing two of those four as short glossary entries of 150 to 250 words each so the term is still covered on your site, and reinvest the freed budget into one UK pricing piece with real figures and dates, two comparison posts that make an actual recommendation, and one process walkthrough tied to a named CMS with screenshots of the real interface. That is eight retained posts plus four new ones, so the same number of published posts, with the two glossary entries sitting outside that count, and a very different asset base.

Restructuring a post so it gets cited, not just absorbed

Open with the direct answer in the first 40 to 60 words, then earn the scroll. People worry this gives the game away. It does not: the reader who wanted only the one-line answer was never going to convert, and the summary layer already has that line. What keeps the other reader on the page is everything a model cannot compress underneath it.

Specificity is the citation currency. Named standards, real UK prices with the date you checked them, a dated example, a screenshot of your own workflow, the error message you actually saw. Vague writing is summarisable by definition, because there is nothing in it that resists paraphrase.

Structure helps machines lift your points cleanly, so write headings the way people phrase the question, keep a short answer block under each one, use a comparison table where the comparison is genuinely tabular, and close with an FAQ that answers real questions rather than keyword variants.

Schema that helps rather than duplicates

Article plus Breadcrumb on every post. FAQ markup only where the questions are visibly on the page, which is a requirement, not a preference. Worth knowing: Google restricted FAQ rich results to well-known government and health sites in August 2023, so FAQ markup is now mostly a clarity and parsing aid rather than a route to extra SERP real estate. Google's own documentation on AI features and your website is clear that there is no special markup that gets you into an Overview.

Then the own goal we see constantly on WordPress: Yoast, Rank Math or a similar plugin is already emitting Article and Breadcrumb JSON-LD, and the publishing tool adds a second copy. Check the page source, or run the URL through Google's Rich Results Test, before you ship anything new.

How to read the damage in Search Console without panicking

The classic Overview pattern looks like this: impressions up, clicks down, average position flat. A site might see its informational blog URLs gain impressions over a quarter while clicks fall and position barely moves, with service pages holding steady and branded queries untouched. That combination points at the summary layer, not a core update. A core update usually moves position. Falling impressions is a third thing entirely, and usually means indexing, seasonality or lost rankings, so check those before blaming generative search.

Segment or you will learn nothing. Site-wide averages hide the shape of the problem, because a commercial page gaining enquiries will mask a glossary section losing half its clicks. Filter pages by path (/blog/ against your service URLs) and queries by intent, including a regex filter on question words to isolate the informational set.

Where generative surfaces are broken out in your account, use them: reporting for AI Mode and related surfaces has been rolling out and the detail available varies, so check what your property actually shows. We walk through that read in reading the AI Mode report in Search Console.

Last thing, and the one teams skip: export a baseline CSV before you change anything, and judge the change over a quarter rather than a fortnight. Two weeks of data after a content change tells you about the weather, not the climate.

Quality standards that matter more now the summary layer exists

Generic AI-written filler is the fastest route to being summarised and forgotten. It is competent, it is grammatical, and it contains nothing a model could not have generated itself, which is precisely why nothing in it survives the compression. Google does not penalise AI content. It ignores content with no information gain, and so does the Overview. Our write-up on what the data actually shows about AI content ranking goes further into that.

Two habits are worth building into the workflow. Score each piece twice, separately: once on SEO value, once on content integrity, and let integrity win when they disagree. A page that survives the summary layer does so on the strength of detail a model could not compress, not on keyword placement. Then put a sceptical sub-editor pass in front of publication, human or otherwise, whose job is to ask what in this article is genuinely new.

Check for cannibalisation before you commission, not after. Two half-answers to the same query help neither, and thin pages are exactly what the summary layer eats first. Build the brand voice from your own existing writing rather than a tone-of-voice prompt: that is where the first-hand texture comes from, and how we approach research, voice and quality explains the mechanics.

A publishing plan a lean UK team can actually run

Plan every brief against the live top 10 for the target query, including whether an Overview fires and which domains it cites. That single check decides the format before a word is written, and it is the step most content calendars skip entirely.

Ramp cadence gradually. Reacting to an Overview scare by shipping twenty posts in a fortnight produces a visible spike of thin content, which is the opposite of the signal you want. Four good posts a month, held steady, beats a panic quarter.

Decide where the distribution happens, because sometimes the click will not. That means an on-brand blog image, a properly sized Open Graph card, and a social post that carries the actual argument rather than a teaser line above a link. If the reader takes your point from LinkedIn and never visits, you still made the point.

Then run a monthly loop: refresh keyword data, re-check Overview presence on your priority queries, and retire or merge the posts that no longer earn their place. Handled this way, AI Overviews and UK SEO stop being a threat to react to and become one more input into a plan you were going to review anyway.

Frequently Asked Questions

What percentage of UK searches show an AI Overview?

Nobody outside Google knows precisely. Published figures come from third-party SERP trackers sampling fixed keyword sets, and they vary by sample, device, country and month. Your own number is the one that matters: export your top 50 UK queries from Search Console and check each one in a clean search.

Do AI Overviews reduce clicks to small business websites?

On informational queries, often yes, and the pattern usually shows as impressions up with clicks down and position flat. Transactional, local and branded searches are much less affected, which is why service-led SMEs frequently see little change in enquiry volume. Measure by page type rather than site-wide, or the two effects cancel each other out in the average.

Which blog topics still earn clicks when an AI Overview appears?

Anything that cannot be compressed into three sentences without losing its value: pricing with real UK figures, comparisons that make a judgement call, step-by-step processes tied to a specific tool or CMS, and pieces built on first-hand experience. Simple definitions and "what is X"explainers are the weakest format now. Keep those as short glossary entries instead of full posts.

How can I tell in Search Console whether AI Overviews are costing me traffic?

Filter to UK, split blog URLs from service URLs, and compare impressions, clicks and average position across a quarter against a saved baseline. Rising impressions with falling clicks and a stable position on informational pages points at the summary layer. Falling impressions usually means something else, such as indexing problems or lost rankings.

Does FAQ or Article schema help a page get cited in AI Overviews?

There is no markup that gets you cited; Google's documentation on AI features is explicit about that. Article and Breadcrumb schema are still worth having for clean parsing, and FAQ markup is fine where the questions genuinely appear on the page, though FAQ rich results have been limited to government and health sites since August 2023. Check your SEO plugin first so you do not publish two copies of the same JSON-LD.