"How do I get ChatGPT to recommend my business" is the newest version of a very old question, and the answer has the same shape as it always did: be the sort of source the system is looking for.

There is no portal, no submission form and nothing to buy. What there is, is a set of things that make you more likely to be retrieved, extracted and cited.

How a recommendation actually happens

When someone asks an assistant to recommend a provider in your category, roughly this occurs:

  1. The question is reformulated into search queries
  2. Pages are retrieved — from a search index, from the model's training, or both
  3. Relevant passages are extracted
  4. An answer is assembled, naming a few sources

You can influence steps one through three. You cannot influence step four directly.

Note what this means: assistants mostly read the same web you already optimise for. This is not a separate channel requiring separate content. It is a different consumer of the same material.

Be crawlable by AI user agents

The most basic requirement and a surprisingly common failure.

Check robots.txt does not block GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot or Google-Extended. Some platforms and security products block these by default, and sites then wonder why they are never mentioned.

Blocking them is a legitimate choice for a publisher whose product is the content. For a services business, being cited is worth considerably more than the marketing copy is.

Write content that can be extracted

A model needs a passage that answers the question and stands alone when lifted.

Answer directly, immediately, under a question-shaped heading. Not after three paragraphs of context.

Make passages self-contained. "This typically costs..." is useless when extracted. "An AI calling agent typically costs..." survives.

Be specific. Models quote checkable statements more readily than impressive-sounding generalities. "Under one second of response latency" gets used; "lightning-fast performance" does not.

That last point is the one most businesses fail, because their content is composed almost entirely of the unquotable kind.

Third-party mentions matter more here

This is where AEO genuinely diverges from traditional SEO.

When someone asks "who are the best AI development companies", a model does not primarily read company websites — every one claims to be the best, so the signal is worthless. It reads sources about companies:

  • Industry directories and listing sites
  • Comparison and roundup articles
  • Forum and community discussions
  • Review platforms
  • Press coverage and interviews
  • Conference and event listings

Being mentioned in those is what gets you into the candidate set. A perfect website that nobody else references is a weak signal.

Practically: get listed in the directories relevant to your sector, participate genuinely where your industry discusses things, and earn coverage where you legitimately can. This is slow, unglamorous and the thing that moves the needle.

Entity consistency

Models build an internal sense of what an organisation is. That gets stronger when the same facts appear consistently everywhere.

Your name, location, founding year, services and category should match across your website, your schema markup, your social profiles, your directory listings and anywhere else you appear.

Where directories hold wrong information — and they frequently do, since much of it is scraped — correcting it is worth the effort. Contradictory data makes a model less confident, and a less confident model names someone else.

sameAs in your Organization schema is how you explicitly connect your site to your verified profiles. It is a few lines and it is the cheapest entity signal available.

Answer the questions people actually ask

Assistants answer questions. Content structured as answers to real questions is what gets retrieved.

Mine your own sources: sales calls, support tickets, the search queries in Search Console. These are phrased how people actually phrase things, which is rarely how marketing departments phrase them.

Then write a page or a section that answers each one directly, with FAQ schema where appropriate.

Add an llms.txt

An emerging convention: a markdown file at /llms.txt summarising what your site covers, with links to the important pages. Cheap to produce, not yet universally consumed, plausibly useful.

Low cost, reasonable expected value. Worth doing, not worth agonising over.

What does not work

Keyword stuffing for AI. Models are substantially better at ignoring this than search engines were.

Claiming to be the best. Every competitor says it. Models weight self-description accordingly, which is to say hardly at all.

Mass-produced thin content. Volume without substance is recognised and discounted.

Buying mentions. Beyond the ethical problem, paid placements in low-quality directories are not sources assistants weight highly.

Expecting it to work quickly. Models are retrained periodically and search indexes update continuously, but entity reputation accumulates over months.

Measuring it

There is no rank tracker for this. Measurement is manual and imperfect:

Keep a fixed question list. Twenty questions a prospect might ask in your category. Run them across ChatGPT, Perplexity, Claude and Google AI Mode monthly. Record whether you appear and in what context.

Watch referral traffic. Visits from chatgpt.com and perplexity.ai show up in analytics and are growing for most sites.

Track branded search. When assistants start recommending you, people search your name. An unexplained rise in branded impressions in Search Console is usually this.

Anyone selling you a precise AEO dashboard is overstating what is currently measurable.

The uncomfortable conclusion

AEO rewards businesses that have something specific and verifiable to say, and that other people reference.

That means the work is largely not optimisation. It is having a genuine position, stating it precisely, being accurate about things competitors are vague about, and earning mentions elsewhere. The technical layer amplifies that; it does not substitute for it.

Which is, on reflection, roughly what good SEO always required.

If you want an assessment of how your business currently appears to answer engines — we run the question list and report what comes back — get in touch.