What has changed on results pages
A results page (SERPThe page of results a search engine shows.) has not been a list of ten links for a long time. Google's in-depth guide to how search works describes the final stage, serving results, as choosing what is most relevant for a query and notes that the features shown on a results page change with the query: a search with local intent shows different elements from a search for images. Featured snippets, question panels, local packs, video carousels and shopping results each take attention and clicks from the ordinary listings on the queries where they appear, and the SERP glossary entry describes how they compete with one another.
AI-generated answers are the latest of these features. The site's SEO hub puts them in that lineage: like featured snippets and question panels before them, they change click behaviour on some queries, and the underlying need, to be the source search engines trust for a topic, is what both classic rankings and AI-generated answers draw on. A Featured SnippetA highlighted extract from one page shown as a direct answer in search results. already showed that a page can supply the answer a searcher reads without receiving the visit; an AI-generated answer extends that pattern.
What is not established by the guidance cited here matters as much. These Google pages do not say on which queries AI answers appear, how they select or attribute their sources, how often searchers click through from them, or how their presence affects the clicks a given listing receives. Third-party claims about all of these circulate widely and change quickly. This guide treats them as unknown, and a team should too until it has its own data from its own queries.
| Question | What the cited guidance supports | What remains uncertain |
|---|---|---|
| Is AI-written content allowed? | Rewarded if helpful, however produced; manipulative automation breaches spam policies | How, or whether, production method is detected |
| Do results pages vary by query? | Yes; the features shown change with the query | Which queries trigger AI answers at any given time |
| What content do systems aim to reward? | Original, people-first, trustworthy content showing experience and expertise | Any ranking factor specific to AI answers |
| How are AI answers built from pages? | Not documented in the pages cited here | Source selection, attribution and click-through |
| What do spam policies cover? | Scaled content abuse, site reputation abuse, expired domain abuse, among others | Thresholds for "many pages" or "little value" |
What Google says about AI-generated content
Google's published guidance on AI-generated content, cited in this page's sources, makes a small number of clear statements, and the AI for content guide summarises them for writers. Its systems aim to reward original, high-quality content that demonstrates the qualities summarised as E-E-A-TExperience, Expertise, Authoritativeness, Trust., however the content is produced. Automation has long been used to produce helpful content, and that is not in itself a problem. Using automation, including AI, to generate content with the primary purpose of manipulating rankings is a violation of its spam policies. Where readers would reasonably wonder how a piece was made, saying so can be useful.
The helpful content documentation adds the self-assessment questions that make those statements practical. It asks whether content provides original information, reporting, research or analysis, whether it adds substantial value beyond what other pages offer, and whether a reader leaves having learned enough to achieve their goal. Its signs of search-engine-first content include producing lots of content on many topics with extensive automation and mainly summarising what others say without adding much value. It also suggests thinking about who created a piece, how it was created and why.
Read together, the pages are neither permission nor prohibition. They describe what is rewarded, and the production method is not on the list. The E-E-A-T guide explains, from Google's documentation and the rater guidelines, that E-E-A-T is not a direct ranking factor or a score but a description of what ranking systems are designed to reward. An AI draft has none of those qualities until a person with real experience shapes and checks it.
Scaled content and spam policies
Google's spam policies describe scaled content abuse as generating many pages primarily to manipulate search rankings rather than to help users, and they are explicit that this applies however the content is created. The site's SEO content hub summarises the same policies: they separately prohibit scaled content abuse and also cover expired domain abuse and site reputation abuse. The policies page states that sites violating them may rank lower or not appear in results at all. The guide to Google's spam policies for marketers walks through every policy on that page, from scaled content to site reputation abuse.
The policy matters for AI because AI removes the cost that used to limit volume. A team that once published a few pieces a month can now publish hundreds, and the temptation is to target every variant of every query with its own page. Pages that differ only in a keyword or a place name, rewrites of other sites' content, and batches published to capture queries rather than answer them are the patterns the policy describes, and a generic page does not become acceptable because a person pressed the publish button.
The boundary is purpose and value, not method. A catalogue that generates distinct product descriptions from its own attribute data is using automation to describe things it sells, and the product descriptions guide covers how to keep that specific. A programme that produces thousands of near-identical location pages with no local substance is in the territory the policy names. Where a plan sits between those, the people-first question in the helpful content documentation applies: would an existing or intended audience find these pages useful if they came to the site directly, rather than through search? Building templated page sets so that each page passes that test is the subject of the guide to programmatic SEO without thin pages.
Writing pages worth citing
Because the cited guidance documents no optimisation specific to AI answers, the defensible approach is the one the guidance does describe. A page worth ranking and a page worth citing are, on everything Google has published here, the same page: original, accurate, clear about who wrote it and why they know, and more useful than what already ranks.
In practice that means content a model could not have produced from what is already on the web. First-hand experience, original data, worked examples from the organisation's own practice, clear definitions and direct answers near the top of the page, with the evidence beneath them. It means accuracy that survives checking, because an answer drawn from a wrong page spreads the error. And it means the authorship and trust signals the E-E-A-T guide describes: a named author with real qualifications, sources for claims, and transparency about who publishes the page and why.
Clear structure helps readers and engines alike. Headings that state what each section answers, a summary near the top, tables for comparisons and plain definitions make a page easier to read and easier to quote accurately. None of this is a tactic for AI answers in particular; it is the content standard the SEO content hub sets for every page, and it is the safest bet when the selection method is not public.
Measuring when clicks change
When a results page adds features above the listings, a page can hold its position and receive fewer clicks. The SEO KPIs and reporting guide builds measurement as a chain, from coverage through visibility, clicks and engagement to outcomes, and reading the chain is what keeps a fall in clicks from being misdiagnosed as a ranking problem. In that guide's terms, visibility without clicks is a snippet or result-feature problem; it calls for a different response from a fall in impressions or position.
The seo-strategy framework draws the same conclusion for AI-generated answers: measurement must track impressions and visibility alongside clicks, because a page can be the source of an answer without earning a visit. That leads to a few concrete habits.
- Segment by intent. Informational and transactional queries meet different results pages and behave differently; averaging them hides both.
- Read click-through against position and features. The Search Console guide explains why click-through must be read against position and against the result features on the page.
- Check results pages by hand. For priority clusters, look at the live results page regularly and record which features appear, because the reports do not show it.
- Annotate. Record known changes to results pages with dates, and say plainly where a movement has no confirmed cause.
- Weight outcomes. Visits that remain may be better qualified; conversions from organic landing pages are the link in the chain that matters to the business.
Do not assume a report separates traffic from AI features unless the tool's own documentation says it does, and do not build forecasts on industry-wide click estimates whose method is unknown.
A garden supplier's informational cluster
Wexhollow Garden Supply, an invented online garden retailer, runs a cluster of 40 how-to guides on pruning and planting. Over two quarters, impressions for the cluster rise 8% and average position holds at about 4, but clicks fall 22% and click-through drops from 5.1% to 3.7%. Transactional queries for the same products are flat. The team checks the 15 highest-impression queries by hand and records an AI-generated answer above the listings on 11 of them, with the guide among the linked sources on 6. Rather than rewrite the cluster, it reports visibility and clicks separately, adds the retailer's own trial results and photographs to the 10 guides with the most impressions, adds a product link to each, and tracks conversions from the cluster's landing pages, which hold steady at 140 a month. The report states that the cause of the click fall is not confirmed and that the answer features are a recorded change, not a measured effect.
What stays the same
Much more has stayed the same than the coverage suggests. Google's guide to how search works still describes crawling, indexing and serving: pages are discovered and fetched, analysed and stored, and chosen for each query. A page that cannot be crawled or indexed cannot appear in results at all. The same guide states that Google does not accept payment to crawl a site more often or rank it higher. The technical work of SEOImproving visibility in unpaid search results., making pages crawlable, indexable and fast, is unchanged.
So is the strategy. Research still identifies the queries, content still has to be the best answer, and authority still decides whose answer is used. Intent still decides format. Links and reputation still signal trust. The seo-strategy framework keeps its five pillars, and what changes is the weight given to measurement and the habit of watching each cluster's results page.
Common mistakes
- Buying tactics for AI answers. The cited guidance documents none; anything sold as one goes beyond it.
- Reading the guidance as permission to scale. AI-written is acceptable; many pages made to rank are not, however they were written.
- Reading it as a ban. The guidance describes no penalty for AI use as such.
- Diagnosing a ranking problem from clicks alone. Check impressions, position and the live results page first.
- Reporting one blended click total. Segment by intent and cluster, and show visibility beside visits.
- Treating third-party estimates as facts. Industry figures on AI answers and clicks have unknown methods and change quickly.
- Rewriting good pages in a panic. If a page still holds its position and its outcomes, the results page changed, not the page.
- Neglecting the basics. Crawling, indexing and helpful content still decide whether a page is seen at all.
