How to get your brand recommended by ChatGPT

ChatGPT doesn't rank pages. It reads a handful of sources, forms an opinion, and names two or three brands. Whether you're one of them depends on what those sources say about you, and there is no ranking dial you can turn.

That's the uncomfortable part. The useful part is that the sources are knowable, and most of them are not your website.

51% of B2B software buyers now start their research in an AI chatbot, up from 29%, and in the same G2 survey 33% ended up buying from a vendor they had never heard of before the AI mentioned it. That is the whole opportunity in one number. Someone is being introduced. It may as well be you.

Where does ChatGPT actually get its answers?

Three places, and they behave completely differently. Knowing which one produced a given answer tells you what to do about it.

The first two are what you're optimising for. And here's the thing most people get backwards: you can't rewrite the training data, but the training data is also where the default opinion lives. When a user asks "what's the best AI visibility tracker" and ChatGPT answers without browsing, it's reciting a consensus it absorbed from thousands of pages. Those pages were mostly not owned by the brands being named.

So the strategy splits cleanly. To influence retrieval, fix your own site. To influence the default opinion, get written about.

GPTBot vs OAI-SearchBot, and why the difference matters

These are two different crawlers with two different jobs, and a lot of sites have quietly blocked the wrong one.

• GPTBot collects data that may be used to train future models. Blocking it protects your content from training; it also removes you from the pool that forms ChatGPT's default opinions.
• OAI-SearchBot builds the index ChatGPT Search reads at answer time. Block this and you are invisible in live, cited answers. This is the expensive mistake.
• ChatGPT-User fetches a page when a user explicitly asks ChatGPT to go look at a link.

If you want to be recommended, allow all three:

# robots.txt
User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

Check what you're currently doing before you assume. Plenty of sites picked up a "block AI scrapers" snippet in 2024 and never revisited it. Also check your CDN: Cloudflare's bot-fight and AI-scraper toggles can block these crawlers at the edge even when your robots.txt is permissive, and nothing in your server logs will look wrong.

What actually moves the needle in ChatGPT?

Third-party mentions. Specifically, being present in the pages ChatGPT retrieves when someone asks a buying question in your category.

When we run scans, the pattern is consistent: the brands that get named in ChatGPT answers are the brands that appear in the roundups, comparison posts, review platforms and forum threads that surface for that query. The model isn't reading your homepage and deciding you're good. It's reading a "12 best X tools" post and repeating it.

Rank the levers by how much they move an answer, and it looks like this:

Two things worth calling out.

Review sites punch above their weight because they are structured, frequently updated, and heavily linked, which makes them cheap for a retrieval system to trust. A G2 category page is a machine-readable list of vendors with ratings attached. That is exactly the shape of thing a model wants when asked to compare tools.

Reddit is genuinely load-bearing in a lot of categories, and it's the one place where doing this badly will actively hurt you. A shill account that gets downvoted into the negatives is a permanent, indexed, retrievable page saying your brand astroturfs. Participate honestly or not at all. We wrote more about this in Reddit and AI search.

How do I make my own pages quotable?

Answer the question in the first two sentences under every heading, in plain declarative prose, with the entity named.

Retrieval systems don't lift your page. They lift a passage. So the unit of optimisation isn't the article, it's the paragraph. A paragraph that reads "Spottlo tracks brand visibility across ChatGPT, Perplexity, Gemini and Google AI Overviews, starting at $39/month" is liftable. A paragraph that reads "we believe in empowering brands to take control of their AI presence" is not liftable, because there's nothing in it.

Concretely:

• Put the answer first. Heading, then answer, then elaboration. Never heading, then a wind-up, then the answer in paragraph four.
• Name yourself. Models resolve entities. Pronouns and "our platform" don't. Say the brand name in the sentence that contains the fact.
• Use numbers and dates. "Starting at $39/month" beats "affordable pricing". "As of July 2026" beats "recently".
• Write comparison content honestly. A page that says "Profound starts at $99/mo but that tier tracks ChatGPT only" is retrievable and correct. A page that says "we're the best" is neither.
• Keep tables. Tables get lifted constantly. A pricing table, a feature matrix, a checklist — these are the easiest things in the world for a model to quote accurately.

A quick before/after on the same fact:

The second column is what shows up in an answer. The first column is what most B2B sites are made of.

How do I know if any of this worked?

You ask ChatGPT the questions your buyers ask, repeatedly, and count how often you're named. There is no console, no impressions report, no equivalent of Search Console.

Start by writing the actual buying questions. Not keywords — questions. For an AI visibility tracker, the real prompt set looks like:

What's the best AI visibility tracking tool for a small SaaS?
How do I track whether ChatGPT mentions my brand?
Cheapest way to monitor brand mentions across ChatGPT and Perplexity?
Profound vs Peec AI vs Otterly — which should I pick?
Is there a tool that tracks ChatGPT, Perplexity and Google AI Overviews together?

Then run them. Manually, in a fresh session with memory off, is fine to start — it will take you an afternoon and tell you more than any amount of theorising. What you're recording per prompt: were you mentioned, in what position, and which competitors were named alongside you.

The catch is that answers are non-deterministic. Ask the same question three times and you can get three different vendor lists. A single manual check is a coin flip, not a measurement, which is why you need repeated runs across a fixed prompt set before the numbers mean anything. That repetition is the boring part, and it's exactly what Spottlo automates: 25 prompts per brand, run weekly across all four engines, with mentions, rank and share of voice tracked over time. There's a free report at spottlo.com if you want a baseline before you commit to anything. How it works has the mechanics.

If you'd rather build the prompt set first, how to build a prompt set covers the method, and AI share of voice explained covers what to do with the numbers once you have them.

Does traffic from ChatGPT actually matter?

Yes, and more than the volume suggests. ChatGPT has around 900 million weekly active users and 50 million paying subscribers, and when OpenAI started putting clickable brand links directly into answers, ChatGPT referral traffic jumped 157.7% week over week.

Volume aside, the quality is the story. Semrush found an AI-search visitor is 4.4x as valuable as an organic search visitor by conversion rate. That tracks with intuition: someone who arrives after ChatGPT specifically recommended you for their specific problem is a very different visitor from someone who clicked the fourth blue link.

So the sessions look small in analytics and convert like demo requests. Don't judge this channel by raw sessions.

What to do next

1. Check your crawlers. Pull up yourdomain.com/robots.txt and confirm OAI-SearchBot isn't blocked. Then check your CDN's bot rules, because that's where the real block usually is.
2. Write 20-25 real buyer questions and run them through ChatGPT with memory off. Record who gets named. This is your baseline and it will probably annoy you.
3. List every roundup and review page that ranks for your category and audit which ones you're missing from. Missing from a "best X tools" post that ChatGPT retrieves is a direct, fixable cause of not being recommended.
4. Rewrite your top five pages answer-first. Heading, then the answer with the brand name and a number in it, then the detail.
5. Re-run the prompt set in 30 days and compare. If you don't want to do it by hand, run a free scan and let it happen weekly.