AEO for import and export: getting named when a buyer asks an AI

AEO, GEO, LLMO, AI SEO. Four acronyms, one job: being the business an assistant names when someone asks it who can supply what you supply. The naming is a marketing problem. The work underneath is the same, and most of it is unglamorous.

A buyer sourcing a new supplier used to open Google, scan ten blue links and click three of them. A growing number now open ChatGPT, Claude or Perplexity and ask a question instead: who can supply food-grade stainless drums into the UK with BRCGS certification? They get back four names. If you are not one of the four, you were never in the running — and you will see nothing in your analytics, because there was no click to lose.

What most of it gets wrong

Selling an SEO retainer with a new cover sheet. Most GEO offers are the same keyword reports as last year with the word AI added. Overlapping with SEO is not the same as being SEO — the unit of competition changed from a ranking position to a quoted passage, and the work has to change with it.

Publishing volume instead of answers. Twenty thin articles a month is a liability, not a strategy. Google's own guidance treats mass-produced content written for machines as scaled content abuse. One page that completely answers a real buyer question outperforms twenty that circle it.

Structured data as a checkbox. Adding Organization schema and stopping is common. What matters is whether the specifications, certifications and terms a buyer actually asks about exist as extractable text on the page — not whether a validator returns green.

Shipping content a crawler cannot read. Many AI crawlers execute little or no JavaScript. If your catalogue only renders after a client-side fetch, the crawler sees an empty page. No amount of content strategy survives that, and almost nobody checks it.

The evidence is unusually clear

Most of this field is assertion. One peer-reviewed study actually measured which interventions change whether a generative engine cites you, and the results are not what most agencies sell.

• Cite your sources
  +40%
• Add statistics
  +37%
• Quote named experts
  +30%
• Authoritative tone
  +25%
• Improve clarity
  +20%
• Use technical terms
  +18%
• Widen vocabulary
  +15%
• Optimise fluency
  +15%
• Stuff keywords
  −10%

Source: Aggarwal et al., “GEO: Generative Engine Optimization”, KDD 2024. Fluency optimisation was reported as a range; the lower bound is shown.

Two things stand out. The top two interventions — citing your sources and adding specific figures — are content decisions, not technical ones. And keyword stuffing does not merely fail to help; it measurably costs you visibility.

How to actually do it

Make the page readable without JavaScript

Everything that matters ships as server-rendered HTML. Verify what crawlers actually receive rather than trusting what your browser shows.

Answer the question in the first two sentences

Each page should lead with a complete, self-contained answer of roughly forty to sixty words, then add depth beneath it. That opening block is the unit an assistant lifts and attributes. If your answer only makes sense after three paragraphs of preamble, it will not be quoted.

Structure the entities, not just the pages

Organization, Product, FAQPage and BreadcrumbList markup, kept consistent with your Companies House record, your certifications and your directory listings. Models cross-check, and they resolve conflicting facts by picking someone else.

Open the door to the crawlers that cite

GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended explicitly allowed in robots.txt, plus an llms.txt file and machine-readable pricing so an agent shortlisting suppliers can read your terms without rendering a page. Blocking these crawlers means those systems cannot cite you at all.

Earn the citation with evidence

Cited sources and specific figures are the two highest-scoring interventions in the research above. Put real numbers and real references into your pages, because vague confidence does not get quoted.

The sceptic's questions

Is this not just SEO with extra steps? It shares foundations — fast, crawlable, well-structured pages help both. It diverges on what winning looks like. SEO competes for a position in a list of links. AEO competes to be the source an assistant quotes, which can happen from page two or three because selection weighs structure and corroboration, not just rank.

Can anyone guarantee you appear in ChatGPT? No, and be sceptical of anyone who says otherwise. No engine exposes a ranking API, retrieval behaviour changes without notice, and citation mixes have shifted overnight before.

How would you even measure it? Partly you cannot, and it is honest to say so. Assistants do not pass referrer data the way search engines do, so some of this lands in analytics as direct traffic. The workable method is to run your priority questions across the major assistants on a fixed schedule and record who gets cited — a manual panel, not a dashboard pretending to precision it does not have.

Our buyers are traditional. Do they really use AI to source? The ones who already know you do not. The risk is the buyer who does not — someone sourcing an unfamiliar product, in an unfamiliar market, who asks an assistant to shortlist four suppliers.

Is it too early to bother? The work is the same work that makes you rank properly in classic search: server-rendered pages, clean structure, real answers to real questions. If answer engines stalled tomorrow you would still be left with a faster, better-organised site. That is what makes it a reasonable bet rather than a gamble.

The research behind this

• Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 — the source for the chart above.
• Google, AI features and your website — Google's own position: no special markup or AI-specific files are required for AI Overviews, and content chunked for machines risks the scaled content abuse policy. Worth reading before you buy anything sold as GEO.
• The llms.txt proposal — the convention for giving AI systems a plain-text summary of a site. Not a ranking signal, and nobody should sell it as one, but trivially cheap to publish.
• Schema.org — the vocabulary behind structured data. Boring, stable, and the closest thing there is to telling a machine what your business actually is.

If you do one thing this quarter, write the twenty questions your sales team answers every week and answer each one properly, on its own page, with a source. Everything else in AEO is plumbing. That part is the product.