When Someone Asks an AI About Your Products or Services, Who Does it Name?

Answer engine optimisation is the work of making a brand the source an AI assistant cites. A growing share of buyers now read an answer instead of clicking a result, and if that answer names three companies and none of them is yours, better rankings will not fix it.

Request a quote

The assistants we optimise for

  • ChatGPT
  • Gemini
  • Perplexity
  • Copilot
  • Claude
AI assistant Illustrative

Who should I use for [what you sell]?

There are a few established options in this space. Your brand is frequently referenced for its approach to the problem, alongside two other providers.

Which suits you depends on the size of your operation and whether you need the work done for you or advice while you run it yourself.

Sources

  • 1yourbrand.com
  • 2a-competitor.com
  • 3another-competitor.com

Illustrative. No agency controls what an assistant says.

Why an Assistant Cites One Source and Not Another

No agency controls what a model says, and anyone claiming otherwise is selling something. But the reasons a model picks a source are knowable, and three of them are work you can actually do.

01Resolution

The model has to know what your brand is

Before an assistant can recommend you, it has to resolve your name to a definite thing: what you do, who you serve, where you operate. Where that information is thin or contradicts itself across the web, the model stays quiet rather than risk saying something wrong.

  • Organisation, Person and Service markup that agree with each other
  • Your details consistent across every profile that mentions you
  • Clear connections between your brand and the subjects it belongs to

How it fails: a company described three different ways across its own site, its profiles and its markup. The model cannot tell which is true, so it names a competitor it can be certain about.

02Extraction

The answer has to be liftable from the page

Assistants quote passages, not pages. Content that circles a subject for four paragraphs before answering gives a model nothing to lift, however well it eventually answers. A direct answer, stated early and plainly, is what gets quoted.

  • Questions as headings, answered in the sentence beneath
  • Definitions stated outright, with no hedging
  • Structure a machine can parse: real headings, lists and tables

How it fails: pages that hedge. "It depends" and "there are many factors" are unusable to a model, so it takes the answer from a source willing to give one.

03Verification

The claim has to hold up in more than one place

A fact stated once, on one page, with nothing supporting it, is a fact a model treats carefully. The same fact agreeing across your pages, your markup and the places that mention you is one it will repeat. Corroboration is what turns a claim into something an assistant is willing to say on your behalf.

  • No contradictions between your own pages
  • Claims traceable to something a model can check
  • Authors, dates and credentials visible rather than implied

How it fails: a site claiming something nothing else supports. The model has no way to confirm it, so the claim is left out of the answer entirely.

What the work actually involves

Every engagement starts by finding out where you stand, because optimising for answers you have never read is guesswork. What follows depends on what that check finds.

First

We ask the assistants about you

We build the set of prompts your buyers actually use, then run them across ChatGPT, Gemini, Perplexity, Copilot and Claude, and record what comes back. Who gets named, who gets cited, and where you appear or do not.

This is done by hand and written up rather than pulled from a dashboard. It is slower, and it means you see the actual answers instead of a percentage nobody can check.

Prompt check

  • "Best [service] for [your market]"Not named
  • "How do I choose a [service] provider"Cited once
  • "[Your brand] reviews"Named
  • "[Competitor] alternatives"Not named

  • Entity cleanup

    Making your brand resolvable. Details corrected and made consistent across the profiles and directories a model reads when working out what you are.

    Fixes: resolution

    • Profile consistency
    • Directory records
    • Brand connections

  • Schema and structured data

    Organisation, Person, Service and FAQ markup that agrees with each other and with the page it sits on, so a model reads one definition rather than three.

    Fixes: resolution and verification

    • Organisation
    • Person
    • Service
    • FAQ

  • Answer-first rewrites

    Existing pages restructured so the answer comes first and the reasoning follows, which is the order a model can lift from and a reader prefers anyway.

    Fixes: extraction

    • Question headings
    • Direct answers
    • Parseable structure

  • Question-led content

    New pages for the questions the check found you absent from, written to answer them properly rather than to rank for a phrase.

    Fixes: extraction and coverage

    • Prompt coverage
    • New pages
    • Comparisons

Changes are applied or documented, depending on access. Where you grant it, we make them and record what changed. Where you would rather keep access in-house, your team receives the exact edits and the markup to add.

Questions about answer engine optimisation

The answers below are the ones we give on calls, including the ones that are not flattering to us.

How do you measure whether it is working?

We re-run the prompt check periodically and compare it against the first one. Same prompts, same assistants, recorded by hand, so you can see the actual answers change rather than a number move.

We do not quote a visibility percentage, because we do not think anyone can yet. Assistants give different answers to the same prompt depending on who asks, when, and what the model saw last. A single figure hides all of that. Reading the answers is slower and tells you more.

Can you guarantee our brand will be named?

No, and nobody can. These are third-party systems that change without notice and answer the same question differently on different days.

What we commit to is the work that makes citation more likely: a brand a model can resolve, answers it can lift, and claims it can verify. Those are the conditions for being cited, and they are the parts you actually control.

Do we need SEO in place before AEO works?

Yes. AEO sits on top of SEO rather than replacing it. Assistants draw heavily on pages that already rank, so a site that cannot be crawled, indexed or found is a site they have little reason to read.

If the foundations are not there, the SEO work comes first. We will say so rather than take on AEO that cannot succeed.

Does this help our Google rankings too?

Yes. Nearly everything that makes a page citable makes it a better search result: a direct answer near the top, questions as headings, consistent facts, markup that describes the page accurately.

The two are converging rather than competing. A page rewritten to be quotable by an assistant usually performs better in normal results as well, which is why we treat them as one piece of work rather than two.

Is AEO the same as GEO?

The terms get used differently across the industry. AEO usually refers to appearing in direct answers, GEO to being cited by generative models. In practice the same signals decide both, so we treat them as one service rather than selling them twice.

Do you need access to our site?

Only if you want us to make the changes. Where you grant access, edits and markup are applied and logged as we go.

Where you would rather keep access in-house, your team receives the exact changes and the markup to add, written so whoever holds access can apply them without interpreting anything.

Is this a project or an ongoing engagement?

Either. A project suits a site that needs its entity signals and key pages fixed once. An ongoing engagement suits a business publishing regularly, where new pages need writing to be citable and the prompt check is worth repeating.

Which fits is settled after the first prompt check, once we both know what the gap actually is.

Which assistants do you work on?

ChatGPT, Google Gemini, Perplexity, Microsoft Copilot and Claude. The prompt check runs across all five, because they do not agree with each other and a brand named by one is often absent from another.

The underlying work is not split by platform. A brand a model can resolve is resolvable to all of them, which is why the effort goes into the signals rather than into chasing each assistant separately.

Find out what they say about you

Send the site and the kind of buyer you want to reach. You get a written scope and a quote back, not a discovery call before anyone will tell you anything.

Request a quote Start with an audit instead
Scroll to Top