A quarter of Google searches now show a generated answer, and those answers cut clicks by 38%. What still works, what no longer does, and the concrete work that keeps you visible on both fronts.
For twenty years the question fitted in one sentence: how do I climb Google's results? That is no longer enough. You can now rank first and lose traffic, because a generated answer sits above you on the screen and answers instead of the click.
The measurements agree. Across a sample of 21.9 million searches, roughly a quarter triggered a generated answer in the first quarter of 2026. A controlled experiment measured a 38% drop in organic clicks on the queries concerned, with the share of searches ending in no click at all rising from 54% to 72% when the answer appears. The top result can shed up to 58% of its clicks.
This is not the end of SEO. It is the end of SEO aimed only at ranking. There are now two fronts to win at once: being ranked, and being cited.
The chain used to be simple: I publish, I rank, I get clicks. It has gained a step. Today: I publish, I rank, a machine reads my page, and it decides whether to cite me or merely summarise me.
The consequence is uncomfortable. Part of your visibility will no longer produce a visit. Your name will appear inside an answer with no link clicked. That is lost traffic but gained recognition, and it shifts the question to ask: less "how many visits?" and more "am I present at the moment someone decides?".
There is a more encouraging consequence too: the traffic that remains is better. The searches a generated answer can satisfy are the purely informational ones. The searches where people want to compare, reassure themselves, buy or get in touch still end in a click. Those are exactly the ones that pay your bills.
The familiar rules still hold. A fast site, indexable, cleanly structured, with content that answers a precise intent and credible inbound links. Nothing that follows replaces those foundations: the models writing the answers draw very largely on pages that already rank well.
A generated answer picks passages, not pages. It favours what is short, self-contained, factual and attributable. A paragraph giving a dated, sourced figure gets quoted; three introductory paragraphs circling the subject do not.
It is a change of unit. You no longer optimise a whole page, you optimise answer blocks inside the page.
The good news is that they overlap heavily. Clarity, structure, accuracy, freshness: those four qualities serve both ranking and citation. A site that does this work seriously wins on both sides, with no separate strategy.
A slow site loses visitors before its content is ever judged. Aim to render the main content in under 2.5 seconds on mobile over an average connection, and above all remove layout shifts: blocks jumping during load drive people away more reliably than a second of waiting.
An up-to-date sitemap, stable readable URLs, a single canonical version per page, no important content locked behind JavaScript. If your page does not read correctly with JavaScript disabled, some crawlers will not read it either.
Mark up what can be marked up: article, local business, product, frequently asked questions, breadcrumbs. Structured data does not mechanically lift you up the results, but it removes ambiguity about what the page contains. A machine that does not have to guess cites more readily.
Check your robots.txt. Blocking assistant crawlers shields your content from reuse, and it also removes you from any answer they produce. That is a trade-off to make deliberately, not a box left to chance by a theme installed three years ago.
Give the answer first, expand afterwards. This rule serves the hurried reader, improves engagement signals, and hands a machine a self-contained extract to reuse. The introduction that sets the scene for three paragraphs is the surest way to be neither read nor cited.
Descriptive headings rather than clever ones. "Price ranges in 2026" gets quoted; "Let's talk money" does not. Every section should be extractable on its own and still make sense outside its context.
"Our solution is high-performing" is worth nothing to an engine or to a reader. "Processing time drops from 4 days to 6 hours" is. Figures, dates, amounts, durations, sources: that is the material generated answers reuse, because it is the only material they can attribute.
Updating the year in the title fools nobody, least of all a model comparing the content. Revisit the figures, cut what has expired, add what has changed. One page seriously reviewed once a year beats ten pages published and abandoned.
If you have an address or a service area, this is your most profitable ground, and the one generated answers erode least: nobody hires a tradesperson from an assistant's answer without checking the profile, the reviews and the opening hours.
Three jobs, in this order: a complete and accurate business profile, a steady flow of reviews, and dedicated pages for each town or district you genuinely cover. That last part only works if the pages are substantial; duplicating one text and swapping the town name is spotted and penalised.
This is the most neglected point and the one that counts most, on both fronts. A model writing an answer leans on what the web says about you, not only on what you say about yourself.
Being listed in a serious directory for your sector, quoted in a local news piece, present on review platforms, referenced by a partner: each of those traces raises the odds that your name comes out when somebody asks an assistant a question. And unlike bought links, those mentions do not lose their value.
Rank tracking alone no longer says much. Four indicators are worth more:
Watch brand queries too: when they rise without traffic rising, it usually means you are being cited without being clicked. That is progress, not failure.
Producing volume instead of substance. Twenty hollow articles published in a month hurt a site. They dilute authority and provide no quotable passage.
Writing for the machine and not the reader. Keyword stuffing has been detected for years; unedited generated text is too, and faster all the time.
Neglecting the foundations to chase the new thing. Optimising for AI answers on a slow, badly indexed site is repainting a facade with no roof.
Confusing visibility with traffic. You will increasingly be seen without being visited. Steering by visits alone will make you cut what is working.
Giving up after three months. SEO shows its effects over six to twelve months. Those who win are rarely the cleverest, they are the ones who did not stop.
No, but its goal has widened. Purely informational traffic is shrinking, commercial-intent traffic is holding up, and a new form of clickless visibility has appeared. The underlying work is unchanged: a healthy site, accurate content, real authority.
It depends on what you sell. A publisher living off page views has a reason to restrict. A company looking for clients almost always benefits from being cited, clicks or not: citation creates preference, and preference brings a brand search later.
Reckon on three to six months for early signals and six to twelve for stable results. The cost depends mostly on the starting point: a technical clean-up is a one-off engagement, while content and authority require continuous effort. The real risk is not overpaying, it is paying for six months and then stopping.
Three actions, in this order, before any strategy.
First check the technical state: mobile speed, indexing, structured data and robots.txt. That is measurable in an hour and it conditions everything else. Then take your three most strategic pages and rewrite their openings so they answer the question within the first hundred words, with at least one verifiable fact per section. Finally, list the places your company should be mentioned and is not: sector directories, review platforms, partners, local press.
None of these three jobs needs an expensive tool, only consistency. They do assume a site capable of carrying them, which brings back a budget question we have covered elsewhere. And because consistency is exactly what runs out when you are already running a business, part of this follow-up is worth automating. That is the work we do at o2code: getting the foundations straight, producing content that stands up, and tooling the follow-up so it does not rest on your memory.
Most projects fail because they start with a tool instead of the bottleneck. Here is the order that holds, what each stage really costs, and how to tell which one applies to you today.
Read full articleThe right question is not which one is best, but which one will cost you least three years from now. What each family charges you once, every month, and on the day you decide to leave.
Read full articleLet's discuss your project and see how we can help you achieve your goals
Contact Us
