AI search optimization, and the new economics of being recommended
Somewhere tonight, a procurement manager in Singapore will type a sentence into a chat window. Not a keyword. A sentence, with a deadline in it and a little anxiety: Which firms can help a mid-sized logistics company cut freight costs in Southeast Asia without switching carriers?
She will receive a paragraph. In that paragraph will be three company names, a line of reasoning for each, and perhaps a link or two she will not click.
She will not see page two of anything. There is no page two. The shelf has room for three products, and the decision about which three was made before she finished typing.
For twenty-five years, businesses fought over a list. They are now fighting over a sentence. Most of them have not noticed the difference, and the difference is where the money is.
The bargain that built the web
Classic search engine optimization rested on an unspoken bargain. Publishers gave search engines their content for free. The engines, in return, sent them people. Ten blue links worked like a row of market stalls: you paid rent in keywords, backlinks and page speed, and in exchange you got foot traffic.
The bargain is fraying, and the fray can now be measured. In a study of the browsing behaviour of 900 U.S. adults, Pew Research Center found that when Google showed an AI summary, users clicked a traditional result in 8% of visits. Without the summary, they clicked in 15%. Users were also more likely to end their browsing session entirely after seeing one. The answer satisfied them. The stall was bypassed.
Gartner saw this early. In 2024 it forecast that traditional search engine volume would fall 25% by 2026 as generative AI tools became substitute answer engines. Whether that exact number lands is a matter of argument. The direction is not.
So the obvious conclusion is that search is dying, and with it the value of being found. The obvious conclusion is wrong, or at least badly incomplete.
The paradox of the smaller crowd
Here is the strange part. The visitors who do arrive from AI platforms appear to be worth far more than the ones who used to arrive from blue links.
Semrush, analysing traffic across many sites, estimated that the average visitor referred by an AI search tool is worth 4.4 times the average traditional organic visitor, measured by conversion rate.
Semrush sells AI visibility software, so the figure deserves a raised eyebrow. But Microsoft Clarity's analysis of more than 1,200 publisher and news sites pointed the same way, with AI referrals converting to sign-ups at many times the rate of traditional search. Methodologies differ. The pattern holds.
Why would fewer people be worth more?
Because of what happens before they arrive. A person who clicked a blue link was still shopping. They opened five tabs, compared, bounced, returned, forgot. A person who arrives from an AI answer has already done the comparing. The machine did it for them, in conversation, and then named you. They are not arriving to evaluate. They are arriving to confirm.
Classic SEO sold attention. AI search sells something closer to a referral.
And this is the room behind the doorway. The visible topic is a technical shift in how queries are processed. The human topic is older than the printing press: whom do we trust to tell us where to go?
From being found to being vouched for
Before search engines, most buying ran on recommendation. You asked the neighbour who had a good plumber. You asked the colleague who had survived the same audit. Reputation travelled from mouth to ear, and it was hard to fake because it was assembled from many independent witnesses.
Search engines replaced the neighbour with an index. The index could be gamed, so an industry grew up to game it: keyword density, link farms, thin pages written for crawlers rather than people. Classic SEO at its worst was the art of appearing relevant without being so.
Large language models restore something like the neighbour. They do not merely index; they synthesize. They read what the web says about you across many sources, weigh it, and compress it into a judgement. When an assistant recommends a firm, it performs a crude version of what a well-connected friend does. It aggregates reputation.
This is why the economics change. You can buy a backlink. It is much harder to buy a consensus.
What AI search optimization actually involves
Strip away the acronyms (GEO, AEO, LLMO; the alphabet keeps growing) and AI search optimization comes down to a handful of disciplines. None is magic. Together they make a business legible to machines that read like skeptical analysts.
Answer first, then elaborate. Language models extract. Pages that state a clear answer in their opening lines, then support it, are easier to quote than pages that circle for three hundred words before arriving. Write the paragraph you would want an assistant to lift whole.
Be one entity, everywhere. Models build a picture of a company from fragments: your site, your LinkedIn page, directory listings, press coverage, reviews. When those fragments disagree about what you do, where you operate or who leads you, the picture blurs, and blurred entities are not recommended. Consistency of name, description and claims across the web is unglamorous and decisive.
Earn mentions you do not control. This is the great inversion. In classic SEO your own site was the main asset. For AI search visibility, third-party sources often carry more weight: industry publications, forums, review platforms, analyst roundups, podcasts with transcripts. The model trusts what others say about you more than what you say about yourself, which is exactly how a sensible person behaves.
Publish something nobody else has. Original data, proprietary benchmarks, documented case studies, a clearly argued point of view. Assistants need sources for specific claims. If you are the origin of a number, you become the citation.
Answer the long tail of real questions. Semrush's data contains a startling detail. When ChatGPT search cited a webpage, that page ranked in position 21 or lower in Google for related queries almost 90% of the time. Separately, only 44.3% of pages ranking in Google's top ten appeared in at least one AI-generated answer. Page one is neither sufficient nor necessary. What matters is whether you answer the precise, conversational, oddly specific question a real buyer asks. Prompts are the new long-tail keywords, only longer, more specific and closer to a purchase.
Keep the technical door open. Structured data, clean HTML, fast pages, and a robots.txt file that does not quietly block the crawlers AI systems rely on. A surprising number of businesses are invisible to assistants because of a configuration line nobody remembers writing.
Measure the new thing, not the old thing. Rankings and sessions undercount AI influence, and many analytics setups still lump AI referrals into "direct" traffic. The useful metric is share of answer: across a defined set of buyer questions, how often are you mentioned, how accurately, and in whose company? That is what good AI visibility services track, prompt by prompt, model by model, month by month.
None of this abolishes classic SEO. Google's own AI features still lean on pages that are crawlable, authoritative and well structured. The foundation remains. What changes is what you build on it.
The economics, plainly
Consider where the money went in classic SEO. Large content budgets to cover every keyword variant. Link acquisition, sometimes bordering on bribery. Constant defence against algorithm updates that could erase a year of work overnight. And at the end, traffic, much of it unqualified, which then had to be converted by a separate and expensive funnel.
The cost structure of AI search optimization differs in four ways.
First, the visitor arrives warmer. If an AI-referred visitor converts at several times the rate of a search visitor, the same budget buys more revenue even at lower volume. Customer acquisition cost falls not because traffic is cheap, but because part of the persuasion has already happened upstream.
Second, the asset compounds. A backlink can be devalued by a single policy change. A reputation built across dozens of independent sources (a cited study, a respected review, a well-answered forum thread, a consistent public record) is harder to dislodge. It accrues like interest rather than depreciating like equipment.
Third, fewer pages, better pages. Answer engines reward depth and clarity over volume. A company that once paid for a hundred thin articles can redirect that spend into ten authoritative ones and the original research behind them.
Fourth, visibility moves upstream of the decision. In blue-link search you competed at the moment of comparison. In AI search you compete at the moment the shortlist is formed. A place on the shortlist is worth more than the best-written page on a list nobody reads.
There are honest caveats. Measurement is immature. Models change behaviour without announcement. Answers differ between ChatGPT, Gemini, Perplexity and Claude, and between one asking and the next. Anyone who promises guaranteed placement inside an AI answer is selling something that does not exist. The return is real but probabilistic, which is to say it behaves the way reputation always has.
Choosing who helps you
This is why the market for help is suddenly crowded. Every agency has rebranded overnight, and choosing an AI SEO agency now requires a sharper test than admiring its own rankings.
A serious AI search optimization company should be able to show you, before any contract is signed, how you currently appear across the major assistants for the questions your buyers actually ask, including where the models get you wrong. It should talk about entities, citations, third-party presence and original content, not only meta tags. Its AI SEO services should include measurement that separates AI referrals from ordinary organic traffic. And a credible AI visibility agency will tell you plainly that some of the work is not technical at all. It is doing things worth being written about.
That last point is the uncomfortable one. Classic SEO allowed mediocre companies to look excellent to a crawler. AI search is less forgiving, because it reads what everyone else says. The best AI visibility services can amplify a strong reputation, correct a distorted one and make a quiet one audible. They cannot manufacture one from nothing, at least not for long.
The shelf, again
Return to the procurement manager and her three names.
The company that ranked first on Google for "freight cost reduction Southeast Asia" for a decade may not be among them. It optimized for a list. Picture instead a smaller, less polished competitor that published a frank study of regional shipping costs two years ago, was quoted in a trade journal, answered questions patiently in an industry forum, and kept its description identical everywhere it appeared. No single act was remarkable. Together they formed a consensus, and the machine, reading that consensus, said its name.
We spent a generation teaching businesses how to be found. The next generation will reward something older and harder: being worth mentioning when you are not in the room.
That was always what reputation meant.
The only thing that has changed is who is listening.
Sources
Comments