How to rank in AI search results: a measurement-first guide

As of November 2026, ranking in AI search results is not a position but a rate: how often an engine mentions your brand, describes it correctly and cites your site inside answers to the questions your buyers actually ask. HubSpot's review of the evidence puts AI Overviews in roughly 48% of tracked Google searches as of early 2026, which is why teams in Singapore are now asking a geo consultant singapore firm to measure something their rank tracker never reported. This guide sets out what the research says moves citations, how to define the metric before you chase it, and the order of work that fits a small marketing team.

Key takeaways:

• AI-referred visitors are a small share of traffic and a large share of signups, so judge AI visibility by downstream conversions, not by session counts.
• A direct answer of 75–150 words at the top of a section is the structural tactic the sources name most often.
• Schema helps machine comprehension, and its effect on ranking is explicitly nil; treat it as plumbing, not a lever.
• Third-party mentions outweigh your own domain as a citation source, which changes where the budget goes.
• Write the denominator first: engine, question version, language, region and sampling cadence.

What the numbers actually say about AI search traffic

Nadia Lim's habit is to start with the definition and the time range, so start here. The idea that AI search traffic is small and therefore not worth work is half true. Ahrefs analysed its own traffic data and found AI search visitors were just 0.5% of total visitors, yet they drove 12.1% of all signups, a conversion rate 23x that of visitors from traditional organic search. Semrush, on a different dataset, put AI search visitors at an average of 4.4x the conversion rate of standard organic visitors.

Those two figures do not measure the same thing, and that is the point. Ahrefs reports one site's internal ratio; Semrush reports an average across sites with its own definition of an AI referral. Both point the same way, and neither belongs in a board deck as "AI traffic converts 23 times better" without the denominator beside it. What a Singapore SME can take from them is narrower and more useful: the pool is small, the intent is high, and the scoreboard should be signups, enquiries or bookings.

The on-page changes that move AI citations

Four changes appear again and again across the sources, and they differ from classic SEO in emphasis rather than in kind.

1. Let retrieval crawlers in. OAI-SearchBot and PerplexityBot feed real-time answers; block them and the brand drops out of those answers even when it ranks well on Google.
2. Lead each section with the answer. State it in one to three plain sentences, aiming for 75–150 words, before the context.
3. Put comparisons in tables. Tables and JSON-LD schema make relationships readable to a machine in a way prose does not.
4. Refresh on a schedule. Pillar pages and cornerstone content every quarter, FAQ sections every 3 months from new customer questions.

The crawler point deserves its own attention in Singapore, where many agencies inherited a blanket "block AI bots" rule from the era when the only concern was training scrapes. Cloudflare reported AI crawlers at 4.2% of all HTML requests on its network, and GPTBot requests grew 305% over a single twelve-month window. A rule that blocks training and retrieval crawlers together removes you from the answers without protecting anything you had not already published.

Why schema is not the lever most teams think it is

Structured data is where SEO habits and answer-engine needs part company, so separate them. A Search Central post that retired a batch of lesser-used structured data types said plainly that such changes "won't affect how pages are ranked". HubSpot makes the same distinction from the other side: schema is confirmed to help eligibility for AI Overviews, while for ChatGPT and Perplexity the visible question-and-answer formatting matters more, because those models read the rendered content.

A practical reading for a Singapore team: implement Article and FAQPage JSON-LD on your most-visited pages, validate it, and move on. The commercial value comes from visible answer structure and from being cited elsewhere, not from the markup. If an agency sells schema as the main lever for AI answers, ask which metric it expects to move and over what period.

How to define an AI visibility metric you can defend

This is the step most teams skip, and it decides whether a report means anything. Define each term before you collect a data point.

Two controls make the numbers comparable: hold the question version, engine, language, region and sampling cadence fixed, and label no-source answers and missing data separately from "not mentioned". A brand absent from an answer and a brand whose data the tool failed to collect are different results; merging them bends the rate in whichever direction the error runs.

A worked example, using Topkee's published reporting format, which is a simulated example and not a client result: a question list run across repeated rounds yields a pool of valid answers under one engine, mode, region, language and question version. An illustrative mention count against that pool gives a mention rate; an illustrative own-site citation count gives a source appearance rate, against an illustrative baseline. Repeated mentions inside one answer count once. The arithmetic is trivial; stating the denominator is not, and it is what lets a month-on-month comparison mean something.

<p align="center"><img src="resource://NAKXU5cgciRWoWt2u2R6nw" alt="GEO brand and source report sample showing 120 valid answer samples, brand mention rate and site citation rate" width="720"></p>

The off-site half of the question

Brands are 6.5x more likely to be cited by AI via third-party sources than via their own domains, according to Airops 2025 as reported by HubSpot. That one ratio reorders the budget conversation. If most routes to citation run through review sites, directories, publications and community threads, then a content programme confined to your own domain covers a minority of the surfaces.

Three moves follow, none of them novel: earn mentions and expert quotes in publications your buyers read, keep your name, address, phone and product descriptions identical everywhere, and run your brand queries in the engines regularly to catch misdescription early. Add a Wikidata entry if you do not have one. Treat this as the same work as a local citation clean-up, aimed at a different audience.

What to do first: a three-month order of work

Run the steps in this order; each one makes the next measurable.

1. Week one: audit robots.txt and server configuration, unblock retrieval crawlers, and confirm indexation in Bing as well as Google.
2. Week two: record a baseline with a fixed list of target questions in ChatGPT, Perplexity, Google AI and Claude, noting citation presence per question.
3. Weeks two to three: add JSON-LD Article and FAQPage to your ten most-visited pages and validate them.
4. Weeks three to four: rewrite the opening of your top pillar pages so each begins with a direct answer.
5. Month two: map existing content to clusters, find the gaps, and fill them rather than starting fresh.
6. Ongoing: original research, expert contributions, and a quarterly refresh cycle for pillar pages.

For a geo marketing singapore programme, the sequence holds with one local addition: check whether competitors in the Singapore market are already cited for your trigger questions before you write a target, because a fresh baseline against a mature competitor changes the goal that is achievable.

Three habits keep a geo consultant singapore engagement honest in this market. Keep one question file and date every version of it, so a comparison never mixes two lists. Separate owned-domain citation from third-party citation in the report, since the two move for different reasons. And hand the business owner a quarterly decision, not a dashboard: which question moved, what was published, and what the next quarter spends money on.

The measurement traps that make AI reports useless

Four traps account for most reports that look impressive and answer nothing.

• Platform "visibility scores" with no stated denominator. A score moving from one number to a slightly higher one says nothing unless questions, engine and rounds are identical.
• Mixing engine results. One engine's sample cannot stand in for the consumer result, and engines whose API provides no web search give no citation data at all.
• Reading a market-wide shift as your own failure. If every brand moved the same week, the engine changed, not your content.
• Confusing being allowed to be read with being cited. A crawler visiting your site and an answer naming it are separate events.

<p align="center"><img src="resource://geo-measurement-decision-diagram" alt="Decision diagram: if volatility is high and your metrics fell, diagnose the market change first; if volatility is calm and your metrics fell, inspect your own content" width="720"></p>

Each trap has one fix: write down what was counted, when, on which engine, against which question version, and keep the raw answers so a later reader can check the arithmetic.

How a GEO service approaches ranking in AI answers

Topkee runs this as a monthly GEO service rather than a rank tracker, monitoring how AI engines mention a brand, describe it and cite sources across an agreed set of buyer questions, then closing the gaps with published knowledge content. Its stated aim is brand mention, information accuracy and source performance inside those samples, not a promise of ranking or recommendation outcomes, and that boundary is worth reading before you buy anything in this category.

The method is ASK, UNDERSTAND, IMPROVE, organised around three buyer stages, choosing a supplier, comparing options and evaluating cooperation, rather than around the brand's own name. Six services sit inside it: prompts tracking, query fan-out analysis, sentiment and brand description checks, competitor comparison under identical questions, AI browsing analysis of retrieval sources and cited pages, and knowledge content writing that answers one buyer question per article. Monthly packages come in three tiers, each sized by a maximum number of simultaneously active tracked questions and a matching number of knowledge articles, scoped to one brand, one primary site, one market and one content language. Each month the deliverables are a monitoring report, a content package and ledger, a prioritised action list split by status, and one review.

The buyer stages are where most AEO tool subscriptions stop short. A dashboard tells you a mention rate fell; the stage framing asks which question the buyer was at, and whether the page that should answer it exists at all. That is the difference between a monitoring invoice and a content decision.

That scoping matters for anyone comparing a geo marketing singapore proposal with a global AEO tool subscription. A tool counts questions across every market; a market-scoped engagement counts the questions Singapore buyers type, in the language they type them, and reports on one primary site. The two are not substitutes, and the cheaper one is not measured the way the other is.

Topkee Media supplies the Singapore context: founded in Hong Kong in 2010, with offices in Singapore, Hong Kong and Los Angeles, and formally entering Singapore in October, roughly two years ago. For a local buyer the practical consequence is that the packages are documented per single market, so a Singapore team should build the question list around local buyer language rather than assume a regional list transfers. Asking a prospective geo consultant singapore partner where Singapore sits in their question design is a fair opening question, and a good answer names the market and the language, not just a headcount of questions.

FAQ

Does ranking in AI search results still depend on Google rankings?

Partly. AI Overviews are drawn from Google's index, so classic organic visibility supplies the candidate pool, but the citation decision inside an answer follows different signals: answer structure, entity clarity and off-site mentions. A page can be cited for a sub-query it does not rank for.

How many AI search visitors should a Singapore SME expect?

Small numbers by share. Ahrefs' own data put AI search visitors at 0.5% of total visitors while they produced 12.1% of signups, so a site with modest traffic may see only a handful of AI-referred sessions a month. Track them in GA4 by source and medium, and judge them on enquiries rather than volume.

Should we block AI crawlers to protect our content?

Not wholesale. You can block training crawlers on proprietary material while leaving retrieval crawlers such as OAI-SearchBot and PerplexityBot open; publishers that blocked AI crawlers via robots.txt lost roughly 7% of weekly traffic within six weeks, according to research from Rutgers Business School and Wharton cited by HubSpot. Decide crawler by crawler, with the business case written down.

Does schema markup improve AI search ranking?

Not ranking directly. Google states that retiring structured data types does not affect how pages are ranked, and HubSpot notes schema's confirmed effect is on eligibility for AI Overviews rather than position. Its value is comprehension: it makes content easier for a machine to extract, while the visible answer structure does the heavier lifting.

How long before AI visibility work shows results?

HubSpot's guidance is that AI search optimisation typically yields initial results within two to three months of implementation, which suits a quarterly refresh cadence for pillar content. Treat the first month as baseline collection, not a results period, and compare like with like.

Next step

Start with the technical and structural quarter above, and read how small businesses can compete with big brands using Google Ads if paid and organic budgets share one pot, or AI-driven strategies for 2026 SEA travel demand for a market where AI answers already steer bookings. A four-layer view, service delivery, AI visibility, traffic and engagement, then business results, keeps the AI numbers from crowding out the commercial ones.

Sources

1. "How to rank in AI search results: Expert best practices", HubSpot Blog, 2026-11-11. https://blog.hubspot.com/marketing/ai-search-ranking
2. "HubSpot AEO — AEO tools to track and improve your AI visibility", HubSpot, accessed 2026-11-11. https://www.hubspot.com/products/aeo
3. "Simplifying the search results page", Google Search Central Blog, 2025-06-12. https://developers.google.com/search/blog/2025/06/simplifying-search-results