Home › SEO Guide › What has changed since Google started answering instead of listing
What has changed since Google started answering instead of listing
For a long time, a search engine returned a list and you chose. Now, for a growing share of searches, you first get a composed answer, with a few cited sources underneath. The change has a direct consequence for your business: the answer to your customer's question may arrive without them ever reaching your website.
Ranked and cited are no longer the same thing
In a list of results, position was everything. In a generated answer there are no positions: you either are or are not among the cited sources. And the sources are not chosen strictly by the order of the list. A page in third place can become a source, and one in first place can be missing, if it gives a vaguer answer to exactly what was asked.
Hence the first practical consequence: the content that wins is the content that answers a precise question, in a passage that stands on its own.
Which is exactly what chapter 8 describes. It is not a new technique, it is the same good writing, now become mandatory.
What Google says needs to be done
Here it is useful to go to the source, because many services with new names have been built around the subject. Google has a page dedicated to optimizing for generative features, and its main message is unusually simple:
You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them.
Google Search Central, guide to optimizing for generative features, accessed 18 August 2026
On the same page, Google explicitly discourages what it calls AEO or GEO tricks, among them artificially chopping text into pieces and creating special files, and recommends instead ordinary optimization, done well.
It is the exact opposite of what a good part of the market is selling right now. Worth keeping in mind before buying an “optimization for artificial intelligence” service.
What has nonetheless been measured to work
There is research too, not just statements. The paper presented at the KDD conference in 2024 by researchers from Princeton and three other institutions tested nine ways of modifying a text and measured how often it gets cited. The results on Perplexity, a real engine:
| Intervention | Effect on visibility |
|---|---|
| Adding quotations | Best result, 22% above the unmodified text |
| Adding statistics | Up to 37% on the perception metric |
| Citing sources | Improvement, best in combination with the others |
| Keyword repetition | 21.9 versus 24.1 for the unmodified text, that is, below the level of not intervening at all |
The authors also note an effect that matters for small businesses: the improvements are largest for pages lower in the ranking. Whoever is already in first place has little to gain; whoever is in fifth place has the most.
The part that cannot be solved from the site
AI assistants frequently cite, as sources, pages that do not belong to the businesses they talk about: comparison articles, community discussions, encyclopedias, independent review sites.
The consequence is uncomfortable and important: a considerable part of your visibility in generated answers is built on sites you do not control. There is no setting on your own site that can make up for a complete absence from the rest of the internet.
What can be done, in order of effort: make sure correct and verifiable information about the business exists where people look for it; take part genuinely in the places where your field is discussed; be present on the review platforms specific to your sector. And if the business is well known enough to have an encyclopedia page, make sure that page is accurate.
To be fully honest: this is the part we cannot help you with from inside the site, and it is, in many fields, the decisive part.
What you can measure, starting this summer
Until recently, the question “do I appear in generated answers?” could only be checked manually. Since 3 June 2026, Google has started showing in Search Console a performance report for generative features, separately for Search and for Discover.
It shows how many times the site's URLs appeared in features such as AI Overviews and AI Mode, and which pages exactly. Two limitations: the rollout is gradual, so you may not have it yet, and the report counts impressions, not visits. In chapter 12 we explained why impressions alone say little.
For the other assistants, ChatGPT or Perplexity, there is no report at all. What remains is the monthly manual check, also described in chapter 12.
What has not changed
It is worth ending here, because it is the most useful conclusion of the chapter. A generative engine still needs to find the page, to be able to read it, to trust whoever wrote it, and to encounter a clear answer to a precise question.
That is, indexing, structure, verifiable identity and good writing. Chapters 3, 4, 7 and 8 of this guide. What has changed is that their absence can no longer be made up for with volume or with age.
Key takeaways
- In generated answers there are no positions, only cited sources. You win differently.
- Google says explicitly that it does not need special files for AI, but ordinary optimization done well.
- Figures and sourced quotations measurably increase the chance of being cited. Keyword repetition lowers it.
- The effect is largest for pages that are not already in first place.
- An important part of visibility is built on sites that do not belong to you.
- Since June 2026 you have, for Google, the first official figures in Search Console.
Sources, accessed 18 August 2026
- Google Search Central, guide to optimizing for generative features: developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Google Search Central, announcement of the performance reports for generative AI, 3 June 2026: developers.google.com/search/blog/2026/06/gen-ai-performance-reports
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande, “GEO: Generative Engine Optimization”, KDD 2024, Barcelona: arxiv.org/abs/2311.09735. The figures quoted are from the table with the results measured on Perplexity.ai.