Skip to content
Deep Quiver
Menu
Talk to a CMO

SEO & AEO

How to get your company into AI answers

Buyers ask ChatGPT, Claude and Perplexity for recommendations before they search. Here is how to structure a site so your company is in the answer.

Jon MackinUpdated 6 min read

A growing share of buyers now open an AI assistant before they open a search engine. They type something like “best fractional CMO firms for a $20M manufacturer in New England” and get three names and a paragraph each. If your company is not one of the three, you were not outranked — you were never considered.

That is a different problem from ranking, and it has a different fix. Here is the work, in the order we do it.

1. Answer the question in the first paragraph

An answer engine does not quote a page. It retrieves passages and assembles them. So the first thing on every page that could plausibly answer a question should be the answer to that question, written to stand on its own in 35 to 60 words.

Self-contained is the operative word. No “as we discussed above”, no pronoun that refers to the headline, no sentence that only makes sense after the hero image. If you lifted that paragraph and pasted it somewhere else, it should still be a complete, true, attributable statement.

The post you are reading has one. So does every page on this site, and the schema that governs this blog refuses to call a post finished without it.

2. Make your headings the questions people actually ask

“Pricing” is a label. “How much does a fractional CMO cost per month?” is a question, and it is the question somebody typed. Headings that mirror real phrasing give a retrieval system something to match against, and they give the reader a working table of contents.

Get the phrasing from real sources, not a keyword tool: sales call recordings, the questions in your inbox, what prospects ask on the first call. If three people asked it out loud this quarter, it deserves an H2.

Then answer directly under the heading, before any setup. One direct sentence, then the nuance.

3. Give the machines the facts in a format they cannot misread

Structured data does not make you the answer. It removes the guessing. JSON-LD lets you state, unambiguously, that this is an organisation, these are its services, this is the author, this is a question and this is its answer, this page was published then and updated now.

The ones worth having:

  • Organization — your name, logo, location, profiles, once, sitewide
  • WebPage and BreadcrumbList — where this page sits
  • FAQPage — one per URL, generated from a real FAQ block that is visible on the page
  • BlogPosting plus Person — who wrote it and when it last changed
  • Service — what you actually sell

Two rules. The markup must match the visible page; inventing an FAQ that only exists in the JSON is the kind of thing that gets markup ignored. And generate it from your content rather than hand-writing it into templates, or it drifts the first time somebody edits a page.

4. Publish an llms.txt

llms.txt is a plain-text index at the root of your domain — one line per page with a link and a one-sentence summary — plus an llms-full.txt holding the full text. It is a proposed convention, not a ratified standard, and no crawler is obliged to read it.

We publish both on every site we build anyway. They are generated from the content collections, so they cannot contradict the site, and they take a build step rather than a maintenance habit. Low cost, unproven benefit, honest framing. Our own are at /llms.txt and /llms-full.txt.

5. Be quotable rather than comprehensive

The 3,000 word “ultimate guide” was built for an era when length was a ranking proxy. A model picking a sentence to cite wants the opposite: something short, specific and checkable.

Quotable looks like this. Name real numbers you can stand behind — our fractional CMO retainers run $7,500 to $15,000 a month, and the engagement starts with a fixed-scope 30-Day Marketing Diagnostic. Name the thing instead of describing it. Say who it is not for. Qualify in the sentence rather than three paragraphs later.

Hedged writing is unquotable writing. “Pricing varies depending on a number of factors” cannot be cited by anything, because it says nothing.

6. Keep your entity consistent everywhere

Models build a picture of your company from more than your website. If your name, services and location are spelled three different ways across your site, LinkedIn, Google Business Profile, directories and old press mentions, you are three weak entities instead of one solid one.

Pick one canonical version of each of these and propagate it:

  • Legal and trading name, with the same punctuation every time
  • The service list, in the same words as your services pages
  • Location — ours is Portsmouth, NH
  • Founder name and bio, identical on the site and on LinkedIn
  • The one-sentence description of what you do

This is unglamorous and it is the part most companies skip. It is also the part you control completely.

7. Check whether you are being cited, and be honest about what you can measure

There is no clean attribution for this. Referral traffic from assistants is partial and inconsistently labelled, a citation inside a chat usually produces no click at all, and the same question asked twice can return different companies. Anybody selling you a precise AI visibility percentage is selling you a modelled guess.

What you can do:

  1. Write down ten questions a real buyer would type, including ones where you expect to lose.
  2. Run them monthly, logged out, in ChatGPT, Claude, Perplexity and Google’s AI results.
  3. Record which companies appear, which sources are cited, and whether anything said about you is wrong.
  4. Watch server logs for assistant crawlers, which tells you whether you are being read.
  5. Track branded search and direct traffic as a blunt proxy for having been named somewhere you cannot see.

Point three matters more than it sounds. The first real finding is usually not absence — it is a model confidently stating something outdated about your pricing or your services, sourced from a page you forgot to update.

What this is worth

This is hygiene, not a growth channel. It will not replace demand generation, and it is not a reason to rewrite your whole site this quarter. The work is a few hours per page at most, it makes pages better for human readers too, and it is cheap insurance against being invisible in a place buyers are increasingly starting.

Start with the one page where a buyer decides: your main service page. Add the direct answer at the top, re-cut the headings into the questions you actually get asked, and check that your name and location read the same there as on LinkedIn. Then book the diagnostic if you want someone to do the other forty pages.

FAQ

Straight answers.

Don't see your question? Talk to a CMO and ask it directly.

Is this different from SEO, or just SEO with a new name?

The underlying work overlaps heavily — clear structure, real expertise, clean markup. What changes is the unit of competition. Classic SEO competes for a ranked position for a page; answer engines assemble a response from passages across several sources, so the thing being selected is a paragraph. That favours short, self-contained, factual writing over long pages that only make sense read top to bottom.

How do I know whether I am being cited?

Ask the models directly and repeatedly. Write down the ten questions a buyer would actually type, run them in ChatGPT, Claude, Perplexity and Google's AI results, and record which companies come back and which sources are linked. Do it monthly from a logged-out session. It is a manual sample, not a measurement, but it tells you whether you appear at all — which is the only question worth answering at the start.

Does llms.txt actually do anything?

Honestly, nobody can prove it moves citations. It is a proposed convention, not a standard that crawlers are obliged to honour. We publish one on every site we build because it costs nothing, it is generated from the content so it cannot go stale, and it doubles as a readable index of what the site contains. That is the whole case for it — cheap and harmless, not proven.

Should I stop writing long articles?

No, but change how they open and how they are sectioned. A 2,000 word piece whose argument only lands if you read all 2,000 words is hard for anything to quote. The same piece with a direct answer up front and self-contained sections under question-shaped headings works for both a reader skimming and a model retrieving.

Everything you need.
Nothing you don't.

Spend 30 minutes with a CMO. You'll leave with a clear read on what's working, what isn't, and what we'd do first, whether or not we work together.