There is a question worth separating from the advertising conversation entirely. When someone asks ChatGPT to recommend a supplier, an agency, a clinic or a piece of software, how does it decide who to name, and can you influence it?
You can. Not by paying. OpenAI's policy is that advertising does not influence the answers ChatGPT gives, and that ads are labelled and kept separate from them. Whatever the model names in the answer, it named for other reasons. Those reasons are largely things you control.
Start with the thing that silently disqualifies you
Before any content strategy, check that OpenAI's crawlers can reach your site. This is the most common and least visible failure, and it disqualifies you from both the paid and the earned side.
OpenAI names two crawlers that matter. OAI-AdsBot validates advertising landing pages and is required if you ever advertise. OAI-SearchBot is the one OpenAI says may help it understand public web content, and it recommends allowing both. Access breaks at four layers, and most audits only look at the first:
- Robots rules. Confirm the relevant paths are allowed, explicitly rather than by assumption.
- Firewall and CDN bot mitigation. Cloudflare, Akamai and similar services block traffic that resembles automation, usually with a 403. This is where well-run sites fail.
- Application-level checks. CAPTCHAs, JavaScript challenges, behavioural analysis and session validation all stop an automated reader even after it has passed the first two layers.
- Rate limiting. Watch for 429 responses in your logs around the times a crawler visited.
Verify by user agent and by provider-level bot verification rather than by short-lived IP observations. OpenAI publishes IP ranges for both crawlers if your security stack requires a stable list.
Write pages an assistant can quote
A model answering a question needs a passage it can lift and attribute without rewriting it into something you did not say. That favours a specific kind of page.
Answer the question in the first paragraph, before the context and the caveats. Use headings that match the question a person would actually type. Keep claims specific and attributable: a number, a date, a named market. Put the substance in text rather than in an image or a chart with no caption. State your prices, your coverage and your terms plainly if you can, because vague pages get paraphrased and specific pages get quoted.
Most of this is good editorial practice. The difference is that a model has no patience for a page that buries the answer under three paragraphs of positioning.
Be legible as an entity, not just as a website
Models build a picture of a business from many sources, not from its homepage alone. Consistency across those sources is what makes you a confident recommendation rather than an uncertain one.
That means the same business name, the same description and the same category everywhere you appear. Structured data on your own pages so the machine reading you does not have to infer what you sell. Presence in the reference sources, directories, trade press and comparison content that get cited in this kind of answer. Where you are named, being named consistently.
Publish your data in a form a machine can read
If your catalogue, prices or coverage only exist inside a PDF or behind a form, they are effectively invisible. Product and pricing data published in machine-readable form is one of the strongest signals available, and one of the least contested, because most competitors have not done it.
Measure it properly
The temptation is to type your own category into ChatGPT once a month and see whether you appear. That is not measurement, because the answers vary by phrasing, by conversation history and over time.
Track at the prompt level. Build a set of the questions your buyers actually ask, run them on a schedule, record who is named and in what order, and treat competitor presence as the benchmark rather than your own presence in isolation. Then track referral traffic from AI assistants separately in your analytics, because it behaves differently from search traffic and converts differently too.
Why this is worth doing now
Two reasons. It compounds, so the gap between starting this quarter and starting next year is larger than it looks. And it has no country restriction, unlike the paid platform, which has opened market by market rather than everywhere at once.
We run this work alongside campaigns on the paid platform. See ChatGPT Ads and AI visibility, or Authority OS for the organic foundation it builds on. Send a brief to start with an audit of whether the crawlers can reach you at all.