AEO (answer engine optimization) and GEO (generative engine optimization) are mostly good SEO under new names. Google's own guidance for AI Overviews and AI Mode says there are no additional requirements and no special optimizations. The parts that are genuinely new are crawler access decisions and measurement, not a separate content discipline.
This article sorts claims into three buckets: what the primary sources support, what is plausible but unproven, and what you can ignore. Last verified: October 2026. Search products change often, so check the linked pages before acting.
What Google says about AI features
Google's page on AI features and your website is blunt. It states that there are no additional requirements to appear in AI Overviews or AI Mode, and no other special optimizations necessary. It says SEO best practices remain relevant, and that you don't need to create new machine readable files, AI text files, or markup to appear in these features.
The same page lists the basics: allow crawling, write helpful content, keep good page experience, keep structured data accurate and matching visible content. Controls such as nosnippet, data-nosnippet, max-snippet and noindex apply to these features too. Traffic from them is reported in Search Console under the Web search type, not a separate report.
That is Google's claim about its own product. It isn't proof that nothing else matters, but it is the only authoritative description of how that product treats your site. Anyone selling a "secret" AI Overviews tactic has to explain why it contradicts this page.
Bucket 1: things that clearly help
These are supported by official guidance, and they were useful before AI search existed.
Crawlability and indexability. If a page can't be fetched and indexed, nothing downstream works. Google needs <a href> links it can follow (link best practices). See our Next.js SEO checklist and JavaScript SEO rendering guides for the technical side.
Clear, direct answers. Put the answer to the question early, then support it. This is good writing, and it also makes a passage easy to quote, whether by a snippet system or a human skimming results.
Original information. Google's helpful content guidance asks whether content provides original information, reporting, research or analysis. A summary of other summaries gives a model (or a reader) no reason to cite you. Your own data, tested procedures, pricing you actually quote and decisions you actually made are the scarce inputs.
Clear entities. Say who you are, what you do, where, and for whom, consistently across your site and profiles. Use the same business name, descriptions and links. Structured data helps when it mirrors what is on the page (see structured data in Next.js). It is a clarity aid, not a ranking switch.
Citations and sourcing. Naming sources, authorship and dates is part of Google's "who, how and why" self-assessment. It also makes your claims checkable, which matters if an AI system summarizes you.
Bucket 2: plausible but unproven
Treat these as experiments, not strategy.
- Passage-level formatting. Short self-contained sections, descriptive headings and tables may be easier for retrieval systems to lift. There's no official confirmation that specific formats get cited more. The downside is low because it's also better for readers.
- Brand mentions across the web. It's reasonable that systems learn associations from mentions. No vendor documents it as a lever, and most claimed correlations come from third-party studies with unclear methods. Do the work that earns mentions (useful tools, honest comparisons), and don't buy them.
- Being cited in ChatGPT or Perplexity. Real, but each product has its own retrieval, and none publishes a ranking recipe. You can measure whether you appear for a set of prompts. You can't reliably engineer it.
- Freshness. Keeping facts current is obviously good. Whether touching a date moves citations is unproven, and faking updates is the kind of thing that erodes trust.
Bucket 3: outright snake oil
- "AI visibility score" guarantees. A number from a vendor's own prompt set isn't a ranking. Ask for the prompts, the sampling method and the variance between runs.
- Mass AI-generated pages. Google's guidance on generative AI content says using AI is fine, but producing many pages with little originality or added value can violate the scaled content abuse policy. It also says to manually fact-check AI output before publishing.
- Hidden text or prompts aimed at models. Text invisible to users but written for crawlers or LLMs is a long-standing spam pattern. There is no documented exemption because a model is the reader.
- Paying for "guaranteed placement" in AI answers. The official guidance reviewed here provides no basis for a vendor to guarantee an organic citation. Ask what placement, product and commercial arrangement the offer actually covers.
- Separate markdown mirrors for bots as a ranking tactic. See llms.txt below.
llms.txt: a proposal, not a signal
llms.txt is a community proposal for a markdown file at your site root that lists LLM-friendly content. It is not a web standard, and the official Google and OpenAI search guidance reviewed here does not establish it as a ranking or inclusion requirement.
Google's AI features page says you don't need AI text files. Documentation platforms do generate these files, and developer tools that read them may find them handy. For a marketing site, the cost is small and the evidence of benefit is absent. Don't prioritise it over fixing broken internal links.
Crawler controls: the part that needs a decision
This is where AI search creates a real choice. Different bots do different jobs, and conflating them leads to mistakes.
| Token | Operator | Purpose | Effect of blocking |
|---|---|---|---|
Googlebot | Crawling for Search, including AI features | Blocks crawling; URLs can remain indexed without their content | |
Google-Extended | Controls use for Gemini training and grounding | Per Google, no effect on Search inclusion or ranking | |
OAI-SearchBot | OpenAI | Surfaces sites in ChatGPT search results | Not shown in ChatGPT search answers (may still appear as links) |
GPTBot | OpenAI | Training generative AI foundation models | Signals content shouldn't be used for training |
ChatGPT-User | OpenAI | User-initiated actions | OpenAI says robots.txt rules may not apply |
Sources: Google common crawlers and OpenAI crawler docs. OpenAI notes changes to robots.txt can take around 24 hours to take effect.
An example that allows ChatGPT search but opts out of training for both vendors:
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Disallow: /
User-agent: Google-Extended
Disallow: /
Whether to do this is a business call. If you want to be discoverable in ChatGPT search, allow OAI-SearchBot. If you object to training use, block GPTBot and Google-Extended. Neither choice changes how Googlebot indexes you, per the sources above. Don't block Googlebot to "opt out of AI Overviews": use nosnippet or related controls and understand they also affect normal snippets.
In Next.js, serve this from app/robots.ts so it's versioned with your code.
A practical order of work
- Confirm important pages are indexable and linked internally.
- Rewrite key pages so the answer comes first and claims are sourced.
- Add original material only you have: real process, real constraints, real numbers you can defend.
- Make a deliberate robots.txt decision per bot.
- Track referrals from AI products in analytics as a separate channel, and treat the numbers as directional.
- Ignore anyone selling a separate "GEO" retainer without showing what they would do that isn't items 1 to 5.
If you want a technical review of crawlability and rendering on a Next.js site, our web team does this work.