SEO for AI Search: One Playbook for Five Surfaces

Most businesses now have five places an AI answer can mention them — Google’s AI Overviews, Google’s AI Mode, ChatGPT, Perplexity and Microsoft Copilot — and the instinctive response is to plan five projects. That is the expensive mistake. The five surfaces share one eligibility gate, one retrieval logic and roughly one set of content requirements. What differs between them is mostly measurement and reach, not the work.

This is the consolidated version: one work programme in five layers, ordered so that each layer only matters if the one below it is solved. It is written for a Singapore business with a normal website and a finite budget, not for an enterprise with a dedicated team.

The five surfaces, and what each is worth in Singapore

Priority should follow reach in your own market, and Singapore’s is unusually lopsided. Google carried 92.46% of Singapore search referrals in July 2026 against Bing’s 3.37% (StatCounter). Assistants are enormously popular here as products — Stanford HAI’s 2026 AI Index puts Singapore near the top of global consumer generative-AI adoption at around 61%, against roughly 53% globally — but as a source of website visits they remain small. Conductor’s 2026 benchmark across 13,770 domains and 3.3 billion sessions put AI referral traffic at 1.08% of sessions, of which 87.4% came from ChatGPT.

The five layers of AI search work, with the Singapore market weighting that should set your priorities.

Layer 0: eligibility, which is where most failures actually happen

Google’s guidance for site owners is blunt and worth memorising before you spend anything: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” Eligibility is simply that “a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements”.

The failures we find most often in audits are self-inflicted and invisible from the front end:

• Snippet suppression. Google names nosnippet, data-nosnippet, max-snippet and noindex as the controls that limit appearance in AI features. A max-snippet value set years ago to discourage scraping now quietly removes you from generated answers.
• Retrieval crawlers blocked by accident. A blanket robots.txt rule written to keep AI companies out usually catches search crawlers too. This is the single most consequential distinction in the whole discipline, so it gets its own table below.
• Content that renders only in JavaScript after user interaction, which is retrievable by nobody.
• Pages not indexed at all. Obvious, routinely missed. Start with our technical SEO basics and the on-page checklist.

On crawlers, the vendors’ own documentation draws a line that most blanket blocking advice ignores. Training crawlers take content to improve models and send nothing back. Retrieval crawlers index content so it can be surfaced and cited in live answers, and those citations are what send referrals. Blocking the second group removes you from AI search entirely.

Decide this deliberately rather than by default. Most Singapore SMEs should allow retrieval crawlers and make their own call on training crawlers; publishers with licensable archives reasonably reach a different conclusion.

Layer 1: retrievability, or getting into the candidate set

Once you are eligible, the question is whether your page is pulled as a candidate. This is where AI search genuinely diverges from classic SEO, because the retrieval is not run against the query the user typed.

Google documents that AI Overviews and AI Mode “may use a ‘query fan-out’ technique — issuing multiple related searches across subtopics”. The consequences show up clearly in the data. Ahrefs analysed 863,000 keyword SERPs and 4 million AI Overview URLs in March 2026 and found that only 38% of cited pages rank in the top 10, with 31.2% at positions 11 to 100 and 31.0% outside the top 100. In their July 2025 run, about 76% of citations were taken from the displayed SERP; eight months later that had fallen to roughly 38%.

Four things follow, and they are the substance of Layer 1:

• Cover the topic, not the keyword. Take each commercially important topic and enumerate the fifteen to twenty questions a buyer needs answered. That set approximates fan-out far better than a keyword list. Our keyword research guide covers building it from real demand rather than intuition.
• Answer at passage level. Give each question its own heading and one self-contained paragraph directly beneath it. Retrieval operates on passages; a paragraph that depends on the two above it to make sense is much harder to lift.
• Name entities plainly. Write “Productivity Solutions Grant (PSG)” rather than “the grant”, and name your city, industry and product explicitly rather than relying on context.
• Date and version your claims. “As of July 2026” inside the sentence is both good practice and a strong extraction signal for anything time-sensitive.

Layer 2: citability, or getting quoted once you are retrieved

Being retrieved is not being cited. The best controlled evidence on what tips that decision is still the KDD 2024 paper that coined the term Generative Engine Optimization. Across roughly 10,000 benchmark queries, the modifications that produced the largest visibility gains were adding relevant statistics, quoting credible named sources, citing sources explicitly and improving fluency — with the best methods landing in the region of 30% to 40% on the paper’s visibility metrics. Keyword stuffing scored below the unmodified baseline.

Two things do not work, and both are widely sold:

• Schema markup as an AI visibility lever. Ahrefs studied 1,885 pages that added JSON-LD between August 2025 and March 2026 against roughly 4,000 matched controls and found no citation lift, with a small negative effect on AI Overviews. Google separately states you do not need new markup or AI text files. Keep schema for rich results, which is where it earns its keep.
• Volume. Publishing more thin pages against more query variations is the exact content most efficiently absorbed by a summary. Fewer, harder pieces is now the cheaper strategy as well as the better one.

The practical test we apply to a draft is simple: pick any paragraph at random, read it out of context, and ask whether it answers a question completely and attributes its evidence. If yes, it is citable. If it needs the section around it, it is not.

Layer 3: off-site presence, the layer most sites skip

This is the layer with the strongest observed relationship to AI visibility, and the one almost nobody budgets for.

Ahrefs measured 75,000 brands against AI Overview brand visibility using Spearman correlations. Branded web mentions scored 0.664 and branded anchors 0.527, well ahead of Domain Rating at 0.326, referring domains at 0.295 and raw backlink counts at 0.218. Around 26% of the brands studied had no AI Overview mentions at all. Ahrefs are explicit that correlation is not causation — strong brands are mentioned more for many reasons — but the ordering is consistent and it is the opposite of a link-building priority list.

Signal strength against AI Overview brand visibility, from Ahrefs’ 75,000-brand correlation study.

The same March 2026 Ahrefs study found something equally practical about media type. Among citations to pages that do not rank in Google’s top 100, 18.2% were YouTube URLs, making up 5.6% of all AI Overview citations, and YouTube grew 34% over six months to become the most-cited domain in AI Overviews. A video that answers your top ten buyer questions is a legitimate AI visibility asset, not a branding indulgence.

For a Singapore business, the workable off-site programme is narrow and unglamorous: get accurately listed and described where your industry is catalogued, earn genuine mentions in local trade and business press, appear on credible third-party “best of” and comparison pages by being genuinely worth listing, keep your Google Business Profile and reviews current, and publish at least some content as video. Our AI local SEO guide covers the local half of this, including the review policies you must not breach while doing it.

Layer 4: measurement, with the gaps named honestly

You need three instruments, and none of them is complete.

Search Console. Google announced Search generative AI performance reports in June 2026, separating AI Overview and AI Mode impressions from standard organic data for the first time. They arrived as a limited rollout covering impressions, pages, countries and devices — with no clicks, no click-through rate and no query breakdown. AI feature appearances were always inside the overall Search performance totals, so your headline numbers do not change.

GA4. The AI Assistants default channel captures referrals from sources like ChatGPT, Gemini, Copilot and Grok, requiring the medium to match ai-assistant exactly alongside a recognised referrer. The trap worth writing on a wall: traffic from Google’s AI Overviews and AI Mode is explicitly excluded from that channel and classified as Organic Search. A meaningful share of assistant referrals also arrive with no referrer and land in Direct. Our guides to the GA4 reports worth checking and attribution models cover how to read this without over-claiming.

A fixed prompt set. Fifty to a hundred questions your buyers genuinely ask, run monthly across the surfaces that matter to you, recording whether you are cited and who is cited instead. It is manual and noisy, because these systems are probabilistic and re-run the same query differently. It is also the only instrument that tells you about competitors.

What not to report: a single-month change in any of the three. And do not put AI impressions at the top of a board report as though they were traffic — that is precisely the pattern our guide to vanity metrics warns about.

A 90-day sequence for a Singapore SME

Ninety days is enough to fix eligibility and content and to see movement in impressions. It is not enough to see a revenue effect, and any agency promising one is guessing. Our guide to how long SEO takes in Singapore sets realistic expectations for the underlying work; the AI surfaces do not run faster than the index does.

What to skip

• llms.txt, for AI search purposes. SE Ranking found roughly 10% adoption across 300,000 domains, and its predictive model became more accurate when the llms.txt variable was removed, meaning it added noise rather than signal. No major AI provider has committed to reading it in production for search, and Google’s own documentation says you do not need AI text files. It does have a genuine use with coding and agent tooling, so publish one if developers consume your documentation. Do not expect a citation from it.
• Rebuilding the site “for AI”. If it is crawlable, fast and indexed, it is eligible. If it is not, that is a technical SEO project with a defined scope, not an AI project.
• A separate AI SEO retainer on top of an SEO retainer. Google states there are no special optimisations required. One programme, one budget. We set out how SEO cost in Singapore is actually built up if you want to sanity-check a quote.
• Reflexively blocking every AI bot. Understandable instinct, wrong default for most SMEs. It removes you from the surfaces that send referrals while doing nothing about the ones that already have your content.

Frequently asked questions

What is AI search optimization?

AI search optimization is the work of making a website eligible for, retrievable by and worth citing in AI-generated answers across Google AI Overviews and AI Mode, ChatGPT, Perplexity and Copilot. In practice it is one programme in five layers: eligibility, retrievability, citability, off-site presence and measurement. Google states there are no additional technical requirements beyond being indexed and snippet-eligible.

Do I need a different strategy for each AI platform?

No. The five main surfaces share one eligibility gate and broadly the same retrieval and citation logic, so one content and technical programme serves all of them. What genuinely differs is crawler access policy, which is set per vendor, and measurement, which is fragmented. Prioritise by reach in your market rather than building separate strategies.

Which AI crawlers should a Singapore business allow?

Allow the retrieval crawlers — OAI-SearchBot, Claude-SearchBot and PerplexityBot — because blocking them removes you from the answers that send referral traffic. Training crawlers such as GPTBot and ClaudeBot are a separate, legitimate business decision. Note that Google-Extended controls Gemini and Vertex AI grounding only and has no effect on Google Search, AI Overviews or AI Mode.

How long does AI search optimisation take to show results?

Expect movement in impressions within about 90 days if eligibility and content work are done properly, and longer for anything resembling a revenue signal. The constraint is the same as classic SEO: re-crawling, re-indexing and building genuine off-site presence all take months. Anyone promising citations in weeks is guessing.

Is llms.txt worth adding to my site?

Not for AI search visibility. SE Ranking’s analysis of 300,000 domains found roughly 10% adoption and that removing the llms.txt variable made its predictive model more accurate, and Google states you do not need AI text files. It is genuinely used by coding and agent tooling, so publish one if developers consume your documentation, but do not expect it to earn citations.

How do I measure AI search visibility in Singapore?

Use Search Console’s generative AI performance reports for AI Overview and AI Mode impressions where you have access, a GA4 segment for AI assistant referrals, and a fixed monthly prompt set of buyer questions to track citation share. Remember that GA4 classifies Google’s own AI surfaces as Organic Search rather than as AI traffic, so the GA4 channel alone will understate Google’s contribution substantially.

Where this leaves you

AI search optimisation is not a new discipline bolted onto SEO. It is SEO with a different centre of gravity: eligibility still gates everything, retrieval now happens against questions nobody typed, citation rewards evidence over keywords, and the off-site layer that most sites never funded turns out to matter more than the link count they did fund.

For a Singapore business the priority order is unambiguous, because 92.46% of search referrals still come through Google. Fix eligibility, rewrite your ten most valuable topics for passage-level extraction, fund the off-site layer, and instrument all three. That programme serves every AI surface at once, and it works whether or not the next twelve months of predictions come true.

Start with the AI SEO guide for Singapore for the wider context, GEO, AEO and LLM SEO compared if the terminology is still in your way, or the AI SEO tools guide if you would rather run the diagnostics yourself. If you want us to do it, the AI SEO service page explains how we work, and the case studies show what the underlying SEO produces.

Written by Adrian Tan, Singapore Digital Marketing. Last updated 5 August 2026.

Related guides

The plumbing side of this playbook is covered separately: llms.txt and AI crawlers sets out which bots actually read your site, why the file does almost nothing for AI search, and the robots.txt rules that decide whether a retrieval crawler can cite you at all.