AI SEO: What the Discipline Actually Is and What Is Hype

Quick answer: AI SEO is the practice of adapting search work to results generated by language models, plus the use of those models inside the work itself. It is not a separate product. Roughly three quarters of it is conventional technical and editorial quality, and the remainder is genuinely new.

Ask five vendors what artificial intelligence has done to search and you will get five different answers, most of them shaped by what that vendor sells. This post is an attempt at a clean definition. It sets out what the discipline actually covers, separates the parts that genuinely changed from the parts that were renamed, and gives you a way to judge a proposal without needing to be technical. It is deliberately not a traffic-impact piece for Singapore SMEs, not a workflow guide for using models in your own research and briefs, and not a technical manual for getting cited by answer engines. Those are three distinct problems and they deserve their own treatment. What follows is the conceptual map that sits underneath all of them. Everything here is written from a Singapore vantage point, with local budgets and a small market in mind. If you want to see how we structure the underlying work, our SEO services page is the reference.

Defining the Term Properly, Because Nobody Else Has

The phrase gets used for at least three different activities, which is why conversations about it go nowhere.

The first meaning is optimising to appear inside generated answers. When a search engine or assistant writes a paragraph in response to a query, it draws on sources. Being one of those sources, and being named, is a visibility outcome that did not exist a few years ago. Work aimed at this is sometimes called answer engine optimisation.

The second meaning is using language models to do search work faster. Clustering keywords, drafting briefs, summarising crawl output, generating schema, writing first drafts. This is an internal efficiency question and has nothing to do with how search results are displayed.

The third meaning is the pure vendor sense, where the phrase is a label applied to an otherwise conventional retainer so it can be priced higher. This third meaning is the source of most of the confusion, and quite a lot of the wasted money.

These are not variations on a theme. They have different objectives, different success measures and different risks. A proposal that does not tell you which of the three it is selling is not a proposal you can evaluate. In our experience the fastest way to find out is to ask what the deliverable is in month one, because the answers diverge immediately: a source visibility audit, a workflow change, or a familiar list of technical fixes with a new heading.

What Genuinely Changed

Being honest about this requires separating mechanism from marketing. Three things are substantively different.

Results now compete with a generated summary for attention. This is the real change and it is a change in demand, not in ranking. Your position can be unchanged while the clicks that position produces fall, because the user’s question was resolved before they reached you. Search work has historically optimised for rank. It now has to optimise for rank and for the residual click value of that rank, which are different things and occasionally in tension.

Retrieval rewards different content properties than ranking does. A ranking system can reward a page holistically. A system assembling an answer needs extractable propositions: clear statements, attributable facts, unambiguous entities. A page can rank well and be nearly useless as source material because everything it says is hedged, buried in narrative, or rendered only after a script runs. Conventional wisdom says write for humans first and machines will follow. That has become only half true, because the machine now needs to lift a sentence out of your page and stand behind it.

Content production costs collapsed, so volume stopped being a moat. When anyone can generate forty competent articles in an afternoon, publishing forty competent articles proves nothing. Whatever is scarce becomes valuable, and what remains scarce is first-hand information: your prices, your data, your process, your results, your local knowledge. This is the change with the largest strategic consequence and the least attention paid to it.

Everything else on the average vendor slide is a restatement of one of these three.

What Is Repackaged

The list of things being sold as new that are not is longer, and worth naming.

Structured data is not new. Marking up your pages so machines understand what they contain has been standard practice for over a decade. It matters more now, which is a change in priority, not a new discipline. Any proposal presenting schema markup as an artificial intelligence innovation is describing 2014 with new vocabulary.

Entity clarity is not new. Being unambiguous about who you are, where you operate and what you sell has always mattered for local search. Consistent business information across your site and your listings was table stakes before any assistant existed.

Crawlability and rendering are not new. A page that a crawler cannot read has always been invisible. The only difference is that there is now one more consumer of the crawl.

Content quality is not new, and this is where the most money is wasted. The properties an answer engine prefers, namely clarity, specificity, accuracy and evidence, are the properties good editors have demanded forever. They are not a technology requirement, they are a standards requirement, and no tool supplies them.

Internal linking and information architecture are not new either, though they are quietly more consequential now. A well organised site tells any consuming system how your topics relate, which page is the authoritative one on a subject, and which pages are supporting detail. That was always a ranking advantage. It is now also a retrieval advantage, because a system deciding which of your pages to draw a statement from uses the same structural cues a crawler always did. Nothing about the technique changed. Only the number of audiences for it did.

And measurement dashboards are being oversold hardest of all. Several vendors now offer a proprietary visibility score claiming to show how often a brand appears inside generated answers. The underlying sampling is rarely disclosed, the numbers are not reproducible, and no two vendors agree. We would not make a budget decision on any of them at SME scale. Clicks, enquiries and revenue remain measurable, comparable and yours.

Here is how the two columns compare in practice, with an honest weighting.

The uncomfortable summary is that a competent traditional retainer already delivered most of the right-hand column. If your current provider had you on solid technical foundations and genuinely specific content, you were doing the majority of this before it had a name. If they had you publishing two thin posts a week, no rebrand will save that approach.

How This Plays Out in a Small Market

Singapore compresses the economics in ways that change the correct answer.

Search volumes here are small, so the cost of chasing informational traffic was always high relative to its value. A keyword with 300 monthly searches nationally, of which a fraction are in-market and a fraction of those are commercially relevant to you, was never a strong foundation. When a generated answer takes a share of those clicks, a marginal tactic becomes a losing one. The strategies that survive are the ones that were already close to the transaction.

Professional and regulated sectors behave differently again. In legal and financial services, users research heavily and then verify before acting, because the stakes are high and the regulatory environment is specific. Generic content in these sectors was already weak, since Singapore rules differ materially from those in the markets most global content is written for. A local firm that states Singapore particulars accurately is both more citable and more persuasive, which is the pattern we work to in law firm SEO engagements and, for the same structural reason, across finance sector SEO.

Trust signals carry disproportionate weight locally because the market is small enough that reputation circulates. Named practitioners, verifiable credentials, real addresses and genuine reviews do more work per unit of effort in Singapore than in a market of a hundred million people. These are also precisely the signals that make a page usable as a source.

And technical debt is common. A large share of Singapore SME sites were built by a web vendor with no search involvement, carry heavy page builders, and render important content late. That was a moderate handicap under conventional ranking. Under retrieval it is a harder ceiling, because content that is not present in the served markup may never be considered at all. This is the least glamorous and most frequently decisive item on the list, and it is what our technical SEO work exists to clear.

How to Judge a Proposal Without Being Technical

You do not need to understand model architecture to evaluate a pitch. Four questions do most of the work.

Ask what is being measured, and what the baseline is. A credible proposal names the metrics before the work starts: clicks segmented by query type, enquiries, not just impressions or an invented visibility score. If the measurement plan is vague, the reporting will be vague by design.

Ask what would be different if you removed the artificial intelligence framing. If the deliverables are a technical audit, content improvements and internal linking, then you are buying an SEO retainer, which may be entirely correct and should be priced as one. The framing should not add a premium.

Ask who writes the content and what they know that a model does not. If the answer is that a model drafts and a junior reviews, the output will be structurally average, because the model is trained on the average of what already exists. The only durable advantage is information the model cannot have: your numbers, your cases, your local particulars.

Ask what they will stop doing. This is the question that separates advisors from order-takers. Adapting to generated answers almost always means retiring content, not just adding it. A proposal that only adds is a proposal that has not analysed anything. We have found that the consolidation half of the work is where most of the gain sits, and it is the half nobody wants to sell because it produces fewer deliverables per invoice.

On price, the Singapore market for a genuine retainer generally runs from around SGD 1,500 a month for a small local site to SGD 5,000 or more for a competitive multi-service business, with one-off audits commonly landing between SGD 2,000 and SGD 6,000 depending on site size. A quoted premium specifically for artificial intelligence capability, on top of that, is the clearest single signal to walk away. Our pricing page shows how we break the scope down, and our about page sets out who does the work.

Field notes: The work that moves results is still recognisably SEO, done with more depth. In our finance case study, an independent CFP-licensed advisory firm in Raffles Place, the programme put author profiles, credentials and the MAS licence on display, fixed the technical basics and grew average content length from 380 to 2,200 words per page. Two flagship guides reached position 1 by Months 7 to 8, and monthly organic leads grew from 3 to 31. None of that needed an artificial intelligence premium, so when a proposal charges extra for one, ask exactly which deliverables are genuinely new and why.

Our Take

The discipline is real but much smaller than its marketing. Strip the vocabulary away and the substantive changes reduce to three: clicks now have to be earned after the ranking, pages have to be extractable as well as readable, and volume has stopped being a defensible strategy. Everything else being sold under the banner is work that competent practitioners were already doing.

Our team treats every proposal that leads with an AI feature list the same way it treats one that leads with a keyword count, as marketing rather than method, and asks what specifically changes about the deliverable. We’ve seen the substance of a genuinely updated SEO programme come down to the same three shifts every time: earned clicks after ranking, extractable pages, and no more volume as a strategy on its own.

That conclusion should be liberating rather than deflating. It means you do not need a new vendor, a new platform or a new budget line. You need the same fundamentals executed with more discipline, plus a genuine shift in what you publish, away from restating the knowable and towards documenting what only you know.

The businesses that get this wrong will spend the next two years buying labels. The ones that get it right will spend the same two years becoming the most specific, best evidenced source in their category in Singapore, which happens to be both the ranking strategy and the citation strategy. If you want to see that logic applied end to end in a regulated sector, the finance SEO results write-up follows a case where depth beat volume decisively.

Frequently Asked Questions

Is AI SEO a real discipline or just a marketing term?

Both, unfortunately, which is why the phrase is so slippery. There is a genuine body of work around adapting to generated answers, measuring click loss, and writing pages that can be extracted from rather than merely read. There is also a large volume of conventional search work being relabelled and repriced. The practical test is whether a proposal names a deliverable that would not have appeared in a good retainer three years ago. If it cannot, you are looking at the marketing version.

Does it replace traditional SEO?

No. It is a layer on top of the same foundations. A site that cannot be crawled, loads slowly, or contradicts itself across pages will fail under both the old model and the new one. If anything the fundamentals became less optional, because retrieval systems have even less tolerance for content that is not present in the served markup. Treat it as a change of emphasis within search work rather than a replacement for it.

What should a Singapore SME pay for this?

The same as a competent retainer, which in this market typically runs from around SGD 1,500 monthly for a small local site up to SGD 5,000 or more for competitive multi-service businesses. One-off audits generally fall between SGD 2,000 and SGD 6,000. A specific premium charged for artificial intelligence capability, layered on top of an otherwise ordinary scope, is not something we have seen justified by the deliverables in any proposal we have reviewed.

How is this different from answer engine optimisation?

Answer engine optimisation is the narrower activity of making your pages usable and citable as sources inside generated answers. The broader term also covers measuring and responding to click displacement, and the internal use of models to speed up research and production. The narrower activity is a component of the broader one. Vendors use the two interchangeably, so it is worth asking which they mean before comparing quotes.

Do I need new tools or software?

Almost certainly not at SME scale. Search Console, a crawler and a rank tracker cover the measurement you need, and the analytical work is mostly segmentation you can do in a spreadsheet. Dedicated tools that track brand mentions inside assistants exist and are interesting, but for a business with a small local market the cost is rarely recoverable. Buy the tool after you have identified a decision it will change, not before.

Has content marketing stopped working?

Content that restates what is already widely published has stopped working, and it was already marginal in a market Singapore’s size. Content carrying first-hand information continues to work well and arguably works better, because scarcity increased when production costs fell. The practical shift is from frequency to substance: fewer pieces, each containing something a model could not have produced without you.

Will my rankings drop because of AI?

Not directly. Generated answers sit alongside or above results rather than reordering them, so your position can be stable while the clicks from that position fall. Confusing the two leads to the wrong remedy, since chasing rankings you already hold will not recover clicks that are being absorbed higher on the page. Check average position and click-through rate separately before deciding what problem you actually have.

Is publishing with AI assistance penalised?

The stated position of search engines is that the method of production is not the issue, quality and usefulness are. In practice the risk is commercial rather than punitive: mass-produced content tends to be generic, generic content is poorly differentiated, and poorly differentiated content underperforms whether or not anything penalises it. Use models to accelerate work you could have done well manually, not to produce work you could not.

What is the biggest mistake Singapore businesses are making here?

Buying a solution before diagnosing a problem. We regularly see businesses commit to a rebranded retainer without ever segmenting their own data to establish whether generated answers have cost them anything commercially meaningful. In several cases the traffic decline they were reacting to came entirely from pages that had never produced an enquiry, which makes the spend pure loss. Diagnose first, in your own account, then decide.

Where should a small business start?

Start with an inventory. List your pages, mark which produce enquiries and which produce only sessions, then check clicks against impressions for each group over a year. That exercise usually reveals both the exposure and the opportunity within an afternoon, and it costs nothing. Only after that does it make sense to talk about consolidating content, strengthening technical foundations, or changing what you publish.