Operator Intelligence from Evolette Locin: Agentic SEO and Technical Dominance, Singapore
1. The Operational Flaw: Buying Markup to Win AI Search
A new genre of SEO pitch has appeared in Singapore. An agency tells a founder that Google’s AI answers and the chat assistants run on a special layer of markup, then sells an “AI optimisation package”: a bundle of schema, a new text file for crawlers, and a monthly report on “AI visibility”. The founder pays for a technical layer that the search engine has said it does not require.
Google’s own documentation on AI features and your website is plain on this point. To appear in AI Overviews or AI Mode, a page needs to be indexed and eligible to show in Search with a snippet. Google also says you do not need to create new machine-readable files, AI text files or markup to appear in these features, and that there is no special schema.org structured data to add. The same page explains that these features may run several related searches across subtopics, a technique it calls query fan-out, and may then surface a wider set of supporting links than a classic search would.
Read that second point carefully, because it is where the money is. If the system fans one question out into several sub-questions, your business is not competing for one keyword. It is being checked against a set of facts: who you are, where you are, what you sell, what it costs, whether people trust you. A business whose facts disagree across its own website, its Google Business Profile and its social pages loses on the facts it cannot confirm. Buying more markup does not fix a record that contradicts itself.
2. The Live Proof-of-Concept: Miyu, Winchester, TAG and PickleChoo
We run several businesses that people search for by name and by need, and the same failure shows up in each. Miyu Omakase is a good example of a venue where the facts are small but unforgiving: which outlet, which counter, how many seats, what the booking rules are, and which name belongs to which restaurant. When a guest asks an assistant where to book a counter dinner, a mismatch between the website, the listing and a third-party directory is enough to push the answer toward a competitor with a cleaner record.
The racket sports side is similar. Winchester Tennis Arena, TAG International Tennis Academy and PickleChoo each sell a different service to a different buyer, across different locations. A parent asking about junior programmes, an adult asking about private coaching, and a player asking where to find a pickleball court are three different sub-questions, and each must resolve to the correct business, the correct venue and the correct contact route. For coaching specifically, the verifiable credentials of our coaches, published on their own profile sites such as coachxt.sg, are part of that record.
We do not publish traffic, citation or revenue numbers for these businesses in a public essay, and we would be wary of any adviser who quotes “AI visibility” percentages without showing the underlying data. What we can say is how the work is done and who answers for it. The identity layer behind it is covered in our earlier piece on why disambiguation beats keyword volume, and the page-level discipline in one intent, one page.
3. The Execution Architecture: A Seven-Step Fact Record
Step 1: Confirm the basics are indexable and snippet-eligible. Before anything else, check in Google Search Console that your key pages are indexed, not blocked by robots rules or noindex tags, and not set to suppress snippets. This is the only hard requirement Google states, and a surprising number of business sites fail it through a leftover setting from a redesign.
Step 2: Write the canonical fact sheet. One document, owned by one person, listing the exact legal and trading name, address, opening hours, phone and WhatsApp number, services, price ranges you are prepared to publish, and the policies that affect a customer’s decision. Every public surface is later made to agree with this sheet.
Step 3: Decompose the buyer’s question into sub-questions. Take the five or six commercial questions a customer asks before buying and break each into the parts a system would need to verify. “Where can I book a private omakase dinner?” becomes: is the venue real and located where it says, is it open, can I book, what are the terms, do people trust it. Use an AI agent to generate the sub-question tree quickly, then have an operator remove the ones that do not match how your customers actually buy.
Step 4: Give each sub-question a visible answer on a page. The answer should be in plain text a person can read, in a section with a clear heading, on the page that owns that intent. Do not hide it in a PDF or an image. Google’s guidance asks that structured data match what is visible on the page, so the visible text comes first.
Step 5: Use standard structured data only to describe what is already visible. Organisation, LocalBusiness and Article markup are fine when they mirror the page. They are not a ranking lever for AI answers, and a mismatch between markup and page text is a liability. Keep the author and publisher fields consistent from post to post, with the same names, the same URLs and the same profile pages.
Step 6: Audit every off-site mention against the fact sheet. Google Business Profile, directories, booking platforms, social bios and partner pages all get compared line by line. Fix the differences at the source. This is dull work, and an agent can run the comparison in minutes across dozens of listings, which is the right use of an agent. Correcting a listing that carries a price or a policy should still be signed off by a person, because a published statement is a statement you may be held to.
Step 7: Re-test monthly with real questions. Ask the assistants and Google the exact questions from Step 3, record whether your business appears and whether the facts shown are correct, and log each error against the surface that caused it. Treat this as a diagnostic for your fact record. Results vary by query, location and day, so look for repeated errors rather than reacting to a single answer.
4. The Asymmetric Advantage: What Agencies and Consultants Cannot Copy
An agency selling an AI package is paid for deliverables, and markup is a deliverable. A fact record is not. It has no impressive artefact to show at the end of the month, and it requires someone to chase a booking platform for a corrected address. That work is unglamorous and does not fit an hours-billing model, so it tends not to be done.
An academic consultant can explain query fan-out well and still never have had to explain to a restaurant owner why a wrong set of opening hours on a delivery app is costing real covers. An operator has been on the receiving end of the wrong answer. We know which listing a guest actually used, which message arrived on WhatsApp afterwards, and which error cost the booking. That feedback loop decides what gets fixed first.
There is also a risk dimension that a marketing agency tends to skip. Facts you publish about price, availability, qualifications and policy are commercial representations. In Singapore, misleading consumers about those matters can create exposure under consumer-protection rules, and a sloppy bulk edit across thirty listings can create it quickly. Our legal and contract background means the fact sheet is reviewed as a document that binds the business, not only as a marketing asset.
5. The WhatsApp Conversion Bridge: Request a Private Portfolio Audit
A quick test: ask an assistant and Google five buying questions about your own business, and write down every fact that is wrong, missing or out of date. If you find more than two, the problem is in the record, not in the markup. In a Private Portfolio Audit we build your fact sheet, compare it against your live surfaces, and rank the corrections by how much booking volume depends on them.
Request a Private Portfolio Audit via WhatsApp.
This article is general commercial commentary, not legal or financial advice. Have published pricing, terms and policy statements reviewed against your own records and applicable Singapore consumer-protection rules before you change them across listings.
