GEO for a small brand: entity, quotable structure, third-party mentions

When someone asks ChatGPT which tool to buy, the answer names two or three brands and the click never happens. Getting named is what GEO, generative engine optimization, is for. Ranking is not the same job any more.

For a small brand there are three levers, and I have never found a fourth that matters. A clear entity, so the engines can describe what you are. Quotable structure, so a paragraph of yours can be lifted without editing. Third-party mentions, so the sites the engines already trust say your name. This guide is what to do for each one, in the order that works when you have no team.

If you have not measured yet, do that first: the 5-question AI visibility audit takes 30 minutes and tells you which of these three levers you are on. The rest of this page assumes you know your score.

What you'll have when you're done

• One entity sentence, written down, live word for word on your site, your profiles and your schema
• Five pages that answer your five buying questions in the first two sentences
• A ranked list of the domains AI engines actually cite for your niche, with a next action against each one
• A 30-day sequence with something concrete in it every week
• A monthly re-measurement so you can tell movement from noise
• [SCREENSHOT: the three-lever worksheet filled in for a real brand]

Before you start

• Your 5 buying questions. If you do not have them, write them first: the questions a customer types the week before they choose, none of them containing your brand name.
• Your visibility score out of 15 from the audit, and the list of sources that got named instead of you.
• Access to edit your own site, your LinkedIn, and your profiles on any directory that lists you.
• A Claude or ChatGPT account for the drafting steps. Free plans are fine.
• Two hours to start. The full sequence runs 30 days.

Lever 1: Entity

An entity, in this context, is the engines' internal answer to "what is this thing?". If they cannot answer that in one sentence, they will not risk naming you in a buying answer. This is the least glamorous lever and it is first for a reason: the other two do very little until it is fixed.

Test whether you have a problem

Fresh chat, no memory, three engines:

What is [your brand name]?

Read the answer like a stranger. There are four outcomes and only one of them is fine:

1. It describes you correctly and specifically. No problem here, move to Lever 2.
2. It hedges: "I do not have specific information about..." You have no entity.
3. It guesses a category: "appears to be a marketing agency". You have a vague entity.
4. It describes a different company with a similar name. You have a collision, and it is the most urgent version of this problem.

Write the sentence

One sentence, three parts: what you are, who it is for, what makes it different. No adjectives that a competitor could also claim.

Before and after, from a real cleanup:

We help ambitious brands unlock their full potential through
data-driven, human-centered growth strategy.

Nothing in that can be indexed. "Ambitious brands" is not a customer, "unlock potential" is not a service, and "data-driven" is claimed by everyone. An engine reading this cannot decide when to name you.

Hongtai Ventures builds AI marketing systems for early-stage AI and
crypto companies: growth automation, content engines and GTM tooling,
run by one operator instead of an agency team.

Now there are handles: a category (AI marketing systems), a customer (early-stage AI and crypto companies), three concrete services, and a differentiator that is falsifiable. When someone asks "who does growth for early-stage AI startups", that sentence can be matched.

Draft yours with help, then cut it yourself:

Here is what my company does, in messy form:
[paste your about page, your pitch, whatever exists]

Write 5 candidate one-sentence descriptions. Rules:
- Structure: [name] does [specific thing] for [specific customer],
  [what makes it different].
- No adjectives a competitor could also use (leading, innovative,
  data-driven, human-centered, best-in-class, cutting-edge).
- Name the actual category a customer would search, not a category
  you invented.
- Under 35 words. Plain sentence, no marketing rhythm.
- Then list, for each candidate, which specific customer question it
  would let an AI answer with my name.

Pick one. Read it out loud. If it sounds slightly boring, it is probably right.

Put it everywhere, word for word

The point is repetition without variation. Same name, same sentence, same spelling:

• Homepage, in the first 200 words of visible text
• About page, first paragraph
• The site's meta description
• LinkedIn company page and your personal LinkedIn About
• Every social bio (shortened, but never rewritten with different nouns)
• Every directory, marketplace or partner page that lists you
• Your Organization schema description field

That last one, minimally:

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Hongtai Ventures",
  "url": "https://jordanhongtai.com",
  "description": "Hongtai Ventures builds AI marketing systems for early-stage AI and crypto companies: growth automation, content engines and GTM tooling, run by one operator instead of an agency team.",
  "sameAs": [
    "https://www.linkedin.com/in/jordanhongtai",
    "https://x.com/TheJordude",
    "https://github.com/hongtaiventures"
  ]
}

sameAs is doing real work: it ties the profiles together so three mentions of the same name resolve to one entity instead of three weak ones.

Lever 2: Quotable structure

The engines are looking for a passage they can lift and attribute. Most marketing pages make that impossible, not because the writing is bad but because the answer arrives in paragraph four.

The shape of a quotable paragraph

Unquotable, and typical:

## Why AI visibility matters in 2026

The search landscape has shifted dramatically over the past two years.
Marketers who built their playbooks around blue links are discovering
that the ground has moved beneath them. In this article, we'll explore
what that means for your brand and share some strategies we've found
effective in our work with clients.

There is no sentence in there an engine can quote as an answer to anything. It is throat-clearing, and it is the default output of most content briefs.

Quotable:

## What is an AI visibility audit?

An AI visibility audit checks whether AI answer engines name your brand
when customers ask the questions that lead to a purchase. You run 5
buying questions through ChatGPT, Perplexity and Google AI Overviews,
log which sources each answer names, and score how often you appear out
of 15 answers.

It takes about 30 minutes and needs no paid tools. The output is a
score and a list of the domains that get cited instead of you.

The differences are mechanical, and you can apply them to any page today:

• The heading is the question, phrased the way a person asks it.
• The first sentence is a definition that stands alone. Delete the heading and it still makes sense, which is exactly what an engine does when it lifts it.
• The second paragraph adds the constraints: time, cost, output. Those are what get quoted next.
• No pronoun with an antecedent above it. "This approach" and "it" are un-liftable, because the referent stays behind on your page.
• Numbers, not adjectives. "About 30 minutes" survives being quoted. "Quick and easy" does not.

Three page shapes that get lifted

Definition-first pages. One question per H2, direct answer underneath, then the detail. Build one for each of your five buying questions.

Comparison tables. Engines quote tables often, and a fair table naming your competitors is more likely to be cited than a page that pretends they do not exist.

| Your tool     | Alternative A | Alternative B
---------------|---------------|---------------|---------------
Starting price | $X/mo         | $Y/mo         | free tier
Best for       | one case      | another case  | another case
Setup time     | 10 minutes    | 2 hours       | 10 minutes
The catch      | your honest   | theirs        | theirs
               | limitation    |               |

Write the "the catch" row honestly, including for yourself. A table that only flatters you reads as marketing and gets skipped; a table that concedes something is the kind of source an engine quotes.

FAQ blocks. Four to six questions at the end of every important page, phrased as people ask them, each answered in two to four self-contained sentences with no "as mentioned above". Add FAQPage schema. Google mostly stopped showing FAQ rich results in 2023, but the AI crawlers still read the markup, and the format itself is the point.

Rewrite an existing page with this prompt:

Here is a page from my site:
[paste the page]

Rewrite it so an AI answer engine could quote it. Rules:
- Turn each section heading into the question a customer would type.
- Under each heading, sentence one is a complete, standalone answer
  that still makes sense with the heading deleted.
- Sentence two or three carries the concrete constraints: time, cost,
  requirements, output.
- No opening sentence may start with "In this article", "The landscape",
  or any variation of throat-clearing.
- No pronoun whose referent is in an earlier paragraph.
- Replace every adjective of degree with a number or delete it.
- Keep my facts. Do not invent numbers. If a number is missing, mark it
  [NEEDS NUMBER] instead of guessing.
- At the end, add 5 FAQ questions phrased the way people ask them, each
  answered in 2 to 4 self-contained sentences.

Lever 3: Third-party mentions

Your own site is one vote. In my audit runs, the sources AI answers actually named were mostly not the brands' own sites: roundups, review platforms, community threads and comparison articles carried far more of the citations. That means the third lever is not on your website at all.

Find the domains that matter for your niche

Do not use a generic list of "high authority sites". Build the list from your own questions.

1. Run each of your 5 buying questions in ChatGPT, Perplexity and Google AI Overviews. Clean context: a temporary chat, a new thread, an incognito window.
2. For each answer, write down every domain cited or linked, in the order it appears.
3. Count. Fifteen answers gives you 40 to 80 domain mentions.
4. Rank by frequency.

Let Claude do the tallying:

Below are 15 AI answers to buying questions in my niche, each under a
heading like "Q1 / chatgpt". Extract every domain that is cited or
linked in each answer.

Output:
1. A table: domain, times cited, which questions, which engines.
   Sort by times cited, descending.
2. Classify each domain: REVIEW SITE / ROUNDUP OR LISTICLE /
   COMMUNITY (forum, Reddit, Discourse) / NEWS OR PUBLICATION /
   BRAND OWN SITE / DOCS / OTHER.
3. For the top 10, one line each on the realistic way a small brand
   gets mentioned there: submit a listing, pitch the author, answer a
   thread, get reviewed, contribute a guest piece, or "no realistic
   path".

Do not include domains that are not in the text below.

ANSWERS:
[paste]

The "no realistic path" verdict is doing useful work. Some citations come from publications you will not get into this year. Naming that honestly stops you burning a month on it.

The realistic ways in, ranked by effort

• Directory and marketplace listings. Lowest effort, and often the fastest citation. If a listing site appears twice in your tally, get listed this week. Use the entity sentence verbatim in the listing.
• Review platforms. Ask your five happiest customers for a review on the platform that appears most in your tally. Not on all of them, on the one the engines cite.
• Community threads. Find the threads the engines cited, and answer the question properly, with your reasoning and one link, in the voice of someone who knows the subject. A thread from 2024 that gets cited today will still be cited next year.
• Roundups and listicles. Email the author with a specific reason you belong in the list: the case they do not cover, the customer segment none of the current entries serve. Reference the actual article. One paragraph.
• Guest contributions. Slowest, and worth it only for the publication at the top of your tally.

Every one of these uses the same entity sentence. That is the compounding: Lever 1 makes Lever 3 legible.

The 30-day sequence

Order matters more than speed. Entity first, because everything else references it.

Week 1: entity. Run the entity test on three engines. Write the sentence. Roll it out to all six or more surfaces including schema. Update every profile. Two focused hours, then done.

Week 2: structure, on five pages. One page per buying question. Heading as the question, answer in the first two sentences, constraints in the second paragraph, FAQ block at the bottom with schema. If the page does not exist, write it; if it exists, rewrite the top of it.

Week 3: the citation list. Fifteen answers, tally the domains, rank them, classify them, write the next action for the top 10. Then execute the cheapest three: listings, a review ask, one community answer.

Week 4: outreach and one comparison page. Two or three roundup pitches, referencing the actual article. Publish one honest comparison page including the "the catch" row about yourself.

Day 30: re-run the audit. Same five questions, same three engines, same sheet, new date. Compare the score, and compare the top-sources list.

How to measure without lying to yourself

Track three numbers, monthly, in the same sheet:

1. Score out of 15. How many of your 15 answers name you.
2. Engine spread. 5 out of 15 concentrated in one engine is a very different position from 5 spread across three, and it tells you whether you have a general problem or an engine-specific one.
3. Repeat sources. Domains cited in both this month's and last month's runs. That is your real target list; anything appearing once is probably noise.

What not to track: how many words you published, how many pitches you sent, whether your rankings moved. Those are inputs. The output is whether the answer says your name.