How to Update Your SEO and Content Strategy for AI Search

Your Search Console graph has a shape you did not have two years ago. Impressions flat or up. Average position flat. Clicks down maybe 20%, maybe 40%, depending on how much of your library is “how to” and “what is”. Nothing broke. You did not get hit by a core update. The queries still fire, you still rank, and fewer people arrive.

So the question on every content lead’s desk right now is a practical one: what do you actually change?

Here is the short version, and the rest of this piece is the evidence and the mechanics. Updating your SEO and content strategy for AI search means moving your content mix away from questions a summary can finish, and toward pages an answer engine has to send someone to: live pricing, real comparisons, original data, first-hand testing, and tools. The second thing you change is measurement, because the referrer data you used to run on is now missing the part you care about. The third thing you do is stop publishing more. Volume is the losing move, and Google’s spam documentation says so in plain words.

What the click data actually shows, and where Google disagrees

There are three decent measurements and one loud rebuttal. Worth knowing all four before you take any of them to a board meeting.

The cleanest is Pew Research Center’s. They instrumented the browsing of 900 US adults from their KnowledgePanel and captured 68,879 real Google searches during March 2025. Of those, 12,593 produced an AI summary. On visits where an AI summary appeared, users clicked a traditional search result on 8% of visits. On visits without one, 15%. Clicks on a source link inside the AI summary: 1% of visits. And 26% of AI-summary visits ended the browsing session entirely, against 16% without.

Second, Ahrefs compared 300,000 keywords, half with an AI Overview present and half informational keywords without one, March 2024 against March 2025. Position-one CTR on AI Overview keywords fell from 0.073 to 0.026. The control set fell too, from 0.056 to 0.031, which tells you some of the decline is not AI Overviews at all. Using the control to forecast, they landed on a roughly 34.5% reduction attributable to the AI Overview itself. The authors flag their own caveat: Search Console gives you no way to isolate AI Overview CTR, so this is a modelled number, not a measured one.

Third, Similarweb data reported by Press Gazette found zero-click news searches on Google rose from 56% in May 2024 to nearly 69% in May 2025, with organic traffic to news publishers falling from a peak above 2.3 billion visits to under 1.7 billion over the same window.

Now the rebuttal. In August 2025 Liz Reid, who runs Google Search, published a post stating that total organic click volume from Google to websites had been “relatively stable year-over-year” and that average click quality had increased. She dismissed third-party reports as based on flawed methodologies or isolated examples.

Both things can be true, and I think they mostly are. Aggregate clicks across the web can hold steady while the distribution underneath rearranges violently. If informational queries stop producing clicks and commercial queries produce slightly more, the total barely moves and your blog still loses 40% of its sessions. Google is reporting the mean. You are living in the variance. Note also that Google published no underlying data, so this is an assertion set against three published methodologies.

The split that decides everything: informational versus commercial

The single most useful number in any of this research is buried in the Ahrefs study: 99.2% of keywords that trigger an AI Overview are informational in intent. Commercial and transactional queries were excluded from their analysis because there was almost nothing to analyse.

Pew found the same pattern from the other direction. Short queries rarely produce summaries (8% of one- and two-word searches) while long ones usually do (53% of ten-word-plus searches, and 60% of question-form queries starting who, what, when or why).

Put those together and you get a clear, uncomfortable map of your own content library. The pages built to catch “what is X”, “how does Y work”, “X vs Y explained at a beginner level”, “benefits of Z” are sitting exactly where the summary lands. The pages built to catch “X pricing”, “buy X”, “X near me”, “book a demo”, “[competitor] alternative” are largely untouched so far.

That is not a reason to delete your explainers. One quoted in an AI Overview still puts your brand name in front of the user, and brand recall is now a distribution channel rather than a soft metric. It is a reason to stop investing new budget at the old rate in the part of the funnel where the click is being intercepted.

Shift the mix toward what a summary cannot finish

The test I use when scoring a proposed piece: can a language model answer this adequately from its training data plus three cited sources, without the user needing to visit anyone? If yes, that piece is a citation play at best. If no, it is still a traffic asset.

Things that consistently fail that test in your favour:

Live pricing, with numbers. Not “pricing depends on your needs”. Real ranges, real units, and what moves the number. Models are cautious about asserting current prices and tend to point the user at a source. Be the only vendor in your category publishing rates and you become that source.
Genuine comparisons, including the case against you. “X vs Y” pages that only flatter X get summarised into nothing. Ones with a real “choose Y if” section get cited, because they contain a judgement the model cannot manufacture.
Original data you generated. A survey of 240 customers. Anonymised benchmarks from your own platform. A teardown of 50 competitor sites. The highest-leverage content type left, because it is the only one that cannot be synthesised from other pages, and it earns the web mentions that turn out to matter.
First-hand testing and operational detail. What broke in month three of the migration. The clause that cost a client money. Google’s helpful-content guidance asks whether content shows expertise “that comes from having actually used a product or service”.
Tools and calculators. An ROI calculator, a sizing tool, a checklist that outputs something. A summary can describe the calculation. It cannot run yours, and someone who wants a number for their own inputs has to arrive.
Anything with a timestamp that matters. Regulatory changes, rate changes, version-specific instructions. Training data ages; live pages do not.

A reasonable reallocation for a mid-sized B2B programme: the budget that produced twelve top-of-funnel explainers a quarter now buys four explainers, one original-data piece, two comparison or pricing pages, and one tool. Fewer URLs, far more expensive each. That is the trade.

What actually gets cited by AI answers

First, the acronyms, without the hype. AEO (answer engine optimisation) and GEO (generative engine optimisation) describe the same practical goal: getting your pages retrieved, quoted and attributed inside an AI-generated answer rather than only ranked in a list of blue links. AIO usually means the same thing with “AI” swapped in. There is no separate algorithm to game. Most of the work overlaps heavily with technical SEO and editorial quality, which is why anyone selling a wholly separate GEO methodology should be asked what it does that SEO does not.

Google’s own documentation is blunt about this. Its page on AI features states there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations needed, and explicitly says you do not need new machine-readable files, “AI text files”, or any special schema.org structured data. If a vendor is selling you an llms.txt file as the fix, that is Google’s answer.

What the data supports:

Ranking still carries most of the weight. Ahrefs analysed 1.9 million citations from one million AI Overviews and found 76.1% of cited pages rank in Google’s top 10 for the query. About 14% did not rank in the top 100 at all, which is the interesting minority, but the base case is clear: classic ranking is the on-ramp.
Brand mentions across the web beat backlink counts. In a study of 75,000 brands across ChatGPT, AI Mode and AI Overviews, Ahrefs found the strongest Spearman correlations with AI visibility were YouTube mentions (around 0.737) and branded web mentions (0.656 to 0.709), with branded search volume at 0.352 to 0.466, Domain Rating at 0.266 to 0.326, and raw backlink count low. Their own caveat, which I will repeat because it gets stripped off every time this is quoted: correlation is not causation, and improving these metrics will not automatically move AI visibility.
Structure that survives extraction. A retrieval system chunks your page. Write so any one chunk stands alone: a declarative answer sentence in the first 40 words under each heading, headings phrased the way a person asks the question, entities named in full rather than as “it” or “the platform”, figures attributed inline.
Schema for what schema is for. Product, Organization, FAQ and Article markup is still worth having for eligibility and entity disambiguation. Treat it as hygiene, not as an AI lever, because Google says it is not one.
Being quoted elsewhere. A statistic of yours cited in fifteen industry articles reaches the model through fifteen doors. This is where digital PR, podcasts, YouTube and analyst mentions stopped being brand fluff and became distribution.

How to measure this when the referrer data is thin

Be honest with stakeholders: measurement here is immature and partially broken, and anyone showing a tidy AI attribution dashboard is showing estimates. Google confirms AI Overviews and AI Mode traffic is folded into the “Web” search type in Search Console with no separate breakdown. Chat interfaces send a trickle of referrers by comparison: Similarweb data reported by Digiday put AI referral traffic to news and media sites at 35.9 million global visits in June 2025, against roughly 11.2 billion from Google search. Growing, nowhere near a replacement.

So you triangulate. Five instruments, none sufficient alone:

Branded search volume. The most under-used signal you already own. If AI answers are exposing people to your name without a click, the effect shows up as growth in branded queries in Search Console. Track branded impressions and clicks as a separate line from non-branded, monthly, and treat it as your awareness proxy.
Direct and dark traffic. Segment direct sessions from new users, excluding your office IPs and known bots. A rising direct-new-user line alongside falling organic is the fingerprint of discovery happening off your site.
Prompt-level share of voice. Build a fixed set of 50 to 150 buying-stage prompts a real customer would type, run them on a schedule across ChatGPT, Perplexity, Gemini and Google AI Mode, and record whether you are mentioned, cited with a link, or absent. The absolute score is noisy. The trend and the competitive gap are usable. Paid tools exist; a scheduled script and a spreadsheet also work.
AI crawler logs. Watch crawl activity from AI user agents against your key URLs. What is being fetched is a step ahead of what is being cited.
Self-reported attribution. Add “how did you hear about us” to your demo form with an explicit “ChatGPT or another AI assistant” option. Unscientific, and currently one of the more honest numbers available.

One warning on goal-setting. Do not put “AI citations” in anyone’s bonus. It is a metric with no stable definition, no independent verification and an obvious incentive to game. Report it, do not compensate on it.

Why publishing more AI-written articles is the losing move

The instinct when clicks fall is to fill the gap with volume, and generative tools make volume nearly free. Google anticipated this and wrote it down.

The spam policy page in Google Search Central defines scaled content abuse as the generation of many pages “for the primary purpose of manipulating search rankings and not helping users”. The first listed example is using generative AI tools to produce many pages without adding value for users. It sits alongside scraping, stitching content from different pages, and creating multiple sites to hide the scaled nature of the content. That is a spam policy, not a ranking suggestion, and violations can cost you manual actions.

Be precise about this, because it gets misquoted in both directions. Google does not prohibit AI-assisted writing; its helpful-content documentation asks whether automation is self-evident to visitors through disclosure, which is a transparency question, not a ban. The policy’s operative words are scaled and without adding value. Twenty thin AI-drafted posts a week chasing keyword coverage sits squarely inside the definition. One heavily edited AI-assisted piece built on interviews and proprietary data does not.

There is a commercial argument too. The pages you can produce at volume with a model are exactly the pages a model already answers without you. You would be spending money to compete with the thing eating your clicks, on its home ground, in the format it does best.

The six shifts, in the order I would make them

1. Audit by intent, not by traffic. Tag every URL informational, commercial or transactional, and check which currently trigger an AI Overview. You now know which half of the library is exposed and which is insulated.
2. Cut publishing volume and raise cost per piece. Half the URLs, triple the budget each. Consolidate overlapping explainers into one strong page rather than refreshing nine weak ones.
3. Commission one original-data asset per quarter. A survey, a benchmark, a teardown. Something with a number in it only you have. That is what earns mentions, and mentions correlate with AI visibility.
4. Build the two or three tools your buyers would actually use. Calculators, configurators, checkers. These are the pages that require arrival.
5. Rewrite your top 30 pages for extraction. Answer sentence under each heading, question-shaped headings, full entity names, inline attribution, current dates, clean schema. A week of work, and the highest-return technical change available.
6. Stand up the measurement stack before you have to defend the budget. Branded versus non-branded split, direct-new-user tracking, a fixed prompt set run monthly, AI crawler logs, form attribution. Get a baseline now; the comparison is worthless without one.

Questions people ask next

Should we block AI crawlers? Only if you sell content access. If you sell anything else, blocking removes you from the answers your buyers are reading while doing nothing about the clicks you already lost. Most B2B and service businesses should stay open and compete for the citation.

Does traffic from AI assistants convert? Volumes are small enough that most sites cannot say with confidence, and the commonly shared conversion multiples come from vendors selling AI visibility tools. Treat them sceptically until you have your own sample. Self-reported form attribution is the fastest way to get one.

How long before this settles? It has not settled and there is no reason to expect it to within the next year. Build for the mechanism rather than the current interface: be the source with something to say that cannot be synthesised, and stay measurable.

If you want help doing this rather than reading about it, that is the work we do. AB7 Solutions runs SEO, AEO, GEO and AIO programmes alongside the content operations behind them, including intent audits, original-research pieces, calculator and tool builds, schema and extraction rewrites, and the prompt-level tracking that tells you whether any of it moved. Talk to us on +1 321 341 7733, or email ab@ab7solutions.com or director@ab7solutions.com. More at www.ab7solutions.com. If the real problem is 400 thin pages and no original data, we will say so first.

Sources: Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results” (22 July 2025; 900 US adults, 68,879 searches, March 2025); Ahrefs, “AI Overviews Reduce Clicks by 34.5%” (300,000 keywords, March 2024 v March 2025); Ahrefs, analysis of 1.9m AI Overview citations (21 July 2025); Ahrefs, “Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews” (75,000 brands, 12 December 2025); Similarweb data via Press Gazette and Digiday; Liz Reid, Google, “AI in Search: driving more queries and higher quality clicks” (6 August 2025); Google Search Central, Spam policies for Google web search, AI features and your website, and Creating helpful, reliable, people-first content.