How to Optimise Your Site for AI Search in 2026

Start with what Google itself says, because it deflates most of the advice you'll read: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimization" (Google Search Central). No magic file, no secret schema, no AI-specific ranking factor. The same helpful, findable content that works for Search is what surfaces in AI answers.

So if there's no special trick, what do you do? Optimising for AI search comes down to three concrete layers, and only one of them is genuinely new. This is the practical, sourced version — what actually moves the needle, and what to ignore.

Key Takeaways

• Google states there are no special requirements for AI Overviews — standard helpful content plus crawlability is the foundation.
• Content layer: the one peer-reviewed study found statistics (+37%) and quotations (+40%) are the biggest levers; keyword stuffing hurt (−9%).
• Structure layer: clean, parseable content and standard schema help machines read you — but FAQ schema no longer earns a Google snippet (that changed in 2023).
• Crawler layer (the genuinely new part): control GPTBot, OAI-SearchBot, and Google-Extended in robots.txt — block them and you're invisible to AI answers regardless of how good your content is.

Layer 1: Content — What Gets You Cited

This is where most of the leverage is, and it's the best-evidenced part of the whole discipline. The only peer-reviewed study on generative-engine visibility tested nine content strategies against a 10,000-query benchmark and measured how each changed a source's presence in AI answers (Aggarwal et al., arXiv:2311.09735):

The instructions write themselves from that ranking:

• Add specific statistics (+37%) — with the source named. A concrete number is what an engine lifts when it needs something to attribute.
• Use attributed quotations (+40%) — the single biggest lever. A named quote is more citable than a paraphrase.
• Cite your own sources (+28%) in-line. Grounded content gets surfaced.
• Write fluently and plainly (+29% for fluency, +14% for readability). Density doesn't help.
• Stop keyword-stuffing (−9%). The old SEO reflex is now a measurable liability.

If you do nothing else, do this layer. It costs nothing beyond writing honestly, it helps human readers who arrive from classic search too, and it's the part with actual evidence behind it.

Layer 2: Structure — Make Yourself Machine-Readable

The second layer is about making your good content easy for a machine to parse — and it's where the most outdated advice lives.

What helps: clean HTML, a clear heading hierarchy, and standard structured data (Organization, Article, Product where relevant) so engines can reliably identify what your page is and who published it. None of this is AI-specific magic; it's the same technical hygiene good SEO always wanted.

What's misunderstood: FAQ schema. It still helps machines parse your question-and-answer content, and question-shaped content is easy for an AI engine to lift — but it no longer earns you the visible FAQ dropdown in Google results. Google restricted FAQ rich results to authoritative government and health sites in August 2023. So add FAQ schema for machine-readability and AI citation, not for a Google snippet Google stopped granting years ago.

What to skip: llms.txt, the Markdown "summary for AI" file every checklist recommends. No major AI engine has confirmed using it, and Google has explicitly said Search does not. It's cheap and harmless to add, but it does not currently drive anything — don't spend real effort on it believing otherwise. We go deeper on both of these in our companion guide, AEO & GEO in 2026 (publishing soon).

Layer 3: Crawler Access — The Part That's Genuinely New

Here's the layer nobody optimised for two years ago, and the one that can silently undo everything above: if the AI crawlers can't read your site, none of your citable content matters.

The engines separate their crawlers by job, and each is controlled independently in your robots.txt. Getting this wrong — or letting a security tool block them by default — leaves you perfectly visible in classic search and completely absent from AI answers.

Crawler	Run by	What it does	Block it and you lose
OAI-SearchBot	OpenAI	Surfaces & links your site in ChatGPT search	Citations in ChatGPT's search answers
GPTBot	OpenAI	Crawls content used for model training	Inclusion in future model knowledge
ChatGPT-User	OpenAI	Fetches a page when a user's prompt requires it	Live retrieval during a user's chat
Google-Extended	Google	Controls use in Gemini/Vertex training & grounding	Grounding in Google's generative models

Sources: OpenAI bots documentation; Google Search Central (Google-Extended). Each is an independent robots.txt control. Note Google-Extended does not affect your Google Search ranking — it only governs generative use.

The strategic decision hiding in that table is one most brands haven't made deliberately: do you want to be a training source, a live-citation source, or both? They're separate switches. You might allow OAI-SearchBot and Google's search crawler (so you're cited in AI answers now) while blocking GPTBot and Google-Extended (so your content isn't absorbed into model training) — or allow everything to maximise presence. There's no universally right answer, but there is a wrong one: never having chosen, and discovering during an audit that a default setting quietly excluded you from the answer layer entirely. Decide it on purpose.

Putting the Three Layers Together

A concrete order of operations, cheapest and highest-impact first:

1. Check crawler access today. Open your robots.txt and confirm you haven't accidentally blocked OAI-SearchBot, GPTBot, PerplexityBot, or Google-Extended — unless you've chosen to. This is a five-minute check that can single-handedly explain invisibility.
2. Fix the content layer. Add specific statistics, attributed quotes, and named sources to your key pages; drop keyword stuffing. Highest-evidence, lowest-cost work there is.
3. Tidy the structure layer. Clean HTML, clear headings, standard schema. Add FAQ schema for parsing, not for a snippet.
4. Skip the noise. No llms.txt faith, no chasing FAQ rich results, no paying for a "5x AI conversion" number that doesn't have a real source behind it.
5. Audit monthly. Ask your real customer questions across ChatGPT, Perplexity, and Google, and watch whether you appear over time.

That's the whole playbook. It's unglamorous precisely because the honest version is — the same specific-and-sourced discipline we apply to every post here, including our data-backed look at why CAC keeps rising.

What This Means in Southeast Asia and India

The regional note is short and worth acting on. Google's AI Overviews reached India in August 2024 and added Indonesian in October 2024, so the AI answer surface these three layers target is already live in your markets and languages. The optimisation content written for SEA and India is close to nonexistent — which means the brand that does this well, in-market, competes against almost no one for the AI answer slot.

Frequently Asked Questions

How do I optimise my website for AI search?

Work three layers. Content: add specific statistics, attributed quotes, and named sources — the tactics the one peer-reviewed study found most effective — and drop keyword stuffing, which measurably hurt. Structure: clean HTML, clear headings, standard schema. Crawler access: make sure AI crawlers like GPTBot, OAI-SearchBot, and Google-Extended can reach your site in robots.txt. Google itself says there are no special requirements beyond helpful, crawlable content.

Does Google require anything special to appear in AI Overviews?

No. Google's documentation states there are no additional requirements to appear in AI Overviews or AI Mode and no special optimization — the same helpful content and standard SEO practices that work for Search apply. What's new is ensuring AI crawlers can access your content and writing in a specifically citable way.

Should I add an llms.txt file?

It's optional and currently low-value. No major AI engine has confirmed it uses llms.txt, and Google has explicitly said Search does not. It's cheap and harmless to add, but it doesn't drive AI visibility today, so don't treat it as a priority over content and crawler access.

What are GPTBot and Google-Extended?

They're crawler controls you set in robots.txt. GPTBot is OpenAI's crawler for content used in model training; OAI-SearchBot surfaces and links your site in ChatGPT's search answers; and Google-Extended controls whether your content is used to train and ground Google's generative models. Each is independent — you can allow live citation while blocking training use, or allow everything. Google-Extended does not affect your Search ranking.

Does FAQ schema still help for AI search?

It helps machines parse your question-and-answer content, which supports AI citation — but it no longer earns the visible FAQ dropdown in Google results, which Google restricted to government and health sites in August 2023. Add it for parsing and AI readability, not for a Google rich snippet.

Is AI search optimisation different in India or Southeast Asia?

The tactics are the same, but the opportunity is larger. Google's AI Overviews launched in India in August 2024 and added Indonesian in October 2024, so the AI answer layer is live in the region's languages — while optimisation content written specifically for SEA and India is scarce, making early movers unusually hard to displace.

Conclusion

"Optimise for AI search" sounds like it should require a new playbook, and Google's own answer is that it mostly doesn't: helpful, crawlable content is the foundation. The genuinely new work is small and specific — write in a citable way (statistics, quotes, sources), keep your structure clean, and make a deliberate choice about which AI crawlers can reach you. Do the crawler check first; it's the fastest way to find out you've been invisible by accident. Everything else is the same honest, specific content that was always the right idea — now with a measurable payoff in the answer layer.

How this post was compiled. Google's "no additional requirements" statement and the preview controls are quoted from Google Search Central's AI-features documentation; the crawler details (OAI-SearchBot, GPTBot, ChatGPT-User, Google-Extended) are verified against OpenAI's and Google's own bot documentation; the content-tactic figures come from the one peer-reviewed study on generative visibility (Aggarwal et al., arXiv:2311.09735, KDD 2024); and the FAQ-rich-result and llms.txt positions are verified against Google's own statements. Where advice commonly given isn't supported by evidence, we say so. Coact is a performance marketing agency working with ecommerce and app businesses across Singapore, India, and Indonesia.

Continue Learning

• AEO & GEO in 2026: A Guide for Southeast Asian Brands (publishing soon) — the full playbook, including what to skip
• Is Your Brand Invisible to ChatGPT? How to Check — audit your visibility before you optimise
• What's a Good ROAS by Industry? — the citable, sourced style in practice
• Talk to our team about a full AI-search optimisation audit for your site.