How to get recommended by ChatGPT

The short version

There is no way to buy or request a mention in ChatGPT’s answers. What is available is the ordinary work: be described accurately in the sources it draws on, answer the buying question plainly on your own pages, and measure whether the share of answers naming you actually moved.

Anyone offering guaranteed placement is guessing. The lists are not published, not stable and not for sale.

Ask it yourself first

Before reading advice, spend ten minutes. Open ChatGPT and ask the three or four questions a customer would ask before they know any company by name:

• Who should I use for [your service] in [your city]?
• Best [your service] near [area] for [specific need]?
• I need [urgent version of your service] today — who can help?

Write down every business it names.

Two things usually surprise people. First, the list is short — three or four names, not ten. Second, ask the same question again and the list changes. That second fact is why a single check tells you almost nothing, and why measurement means repetition rather than a screenshot.

A third thing surprises people less often but matters more: at least one name will be a business you had not been watching, and at least one competitor you worry about will be absent.

How does ChatGPT decide who to name?

Nobody outside OpenAI knows the full answer, and anyone claiming otherwise is guessing. What is observable from the outside:

It draws on what is written about you across the open web, not only on your own site. In one local market we measured, a single assistant answer rested on eleven pages on average — the business’s own pages alongside directories, review sites, local roundups and trade listings. Yours are the ones that set the facts; the others repeat them, well or badly.

It cross-checks. Where your own site, three directories and a review platform describe your services differently, there is less reason to rely on any of them.

It prefers specifics. A source that states services, prices and hours is more useful for answering a buying question than one that says “leading provider”.

It hedges in some categories. Health, legal and financial questions get more cautious answers, sometimes with no names at all. That is a property of the category, not of your marketing.

It does not answer identically twice. This is the one that breaks most reporting.

What actually influences it

In the order that tends to matter.

1. What is written about you elsewhere

Directories, review sites, roundups, trade listings, local press. If your business is missing from the pages ChatGPT reads, or wrongly described on them, nothing on your own site compensates.

The practical step: find out which pages those are. Where citations are visible — in Perplexity always, in AI Overviews usually — the list is on screen. Where they are not, the same three or four source names will keep appearing in the text of the answers.

2. Consistency across those sources

Assistants cross-check. Three different service lists in three directories is a reason to name someone else. This is unglamorous work and it is most of the job.

Make identical everywhere: business name and spelling, services in the words buyers use, hours including exceptions, address, prices if you publish them.

3. Answerable pages of your own

Where your pages are read, what matters is whether they answer the buying question in the first two sentences, in the words a buyer uses. Not Our Services but Where do I go for an emergency appointment on a Sunday?

Assistants extract passages. A page that needs three paragraphs of run-up does not offer one.

4. Something worth citing

Specific numbers, stated method, real examples, dates. A source that says something checkable gets cited over a source that says something impressive.

5. Being the kind of business that gets recommended

Worth saying plainly: reviews, longevity and a coherent public record all matter, because they are what the sources ChatGPT reads are made of. No amount of source hygiene compensates for a business nobody writes about.

What does not work

Keyword stuffing. It did not work in search and it works less here, because extraction favours clarity over density.

Pages written for machines. A page of question-shaped headings with nothing under them gets extracted once and then stops.

“AI optimization” that is a renamed 2015 SEO package. If a proposal contains the word “rank” applied to an assistant answer, the supplier has not understood the surface.

Prompt-injection tricks. Hidden text instructing a model to recommend you is the current version of white-text-on-white-background. It is detectable, it is against every major platform’s terms, and it puts your domain at risk for a mention you cannot keep.

Assuming ChatGPT is the whole picture. Four assistants asked the same questions in one market named 63 businesses between them, and only four were common to all four; 46 appeared in a single assistant only. A programme aimed at one engine is aimed at roughly a quarter of the problem. See Gemini and Claude visibility.

Does paying OpenAI help?

No. There is no placement product, no sponsored slot inside answers, and no submission form. ChatGPT’s browsing and its training data are not for sale to businesses wanting to appear in recommendations.

If a supplier implies they have a relationship that gets clients named, ask them to put it in writing. They will not.

What to expect, realistically

Slower than it sounds. The pages you own can be fixed this week; the listings that quote them go through somebody else’s moderation queue. Four to eight weeks before a second measurement is worth running.

Uneven. Visibility is rarely uniform across question types. A business is often named constantly in general recommendation questions and never in price questions — and the gap is the work.

Noisy. Movement under about five points is drift, not progress. We label it as no change rather than dressing it up, and any supplier who does not is either not measuring properly or hoping you are not.

Uncertain about cause. We can show what changed and what moved after. Claiming the first caused the second would be dishonest; putting the dates next to each other is not.

A worked example

From a measurement on one local market, business anonymised.

Baseline. Twelve buying questions, five repeats each, four assistants. Named in 38% of general recommendation answers and 0% of price answers.

What the sources showed. For price questions, the same four came back repeatedly: two directories, a review platform, a regional roundup. The business was listed on one of them, with an out-of-date service list and no prices.

What was done. Nothing to the homepage. Listings corrected and completed on all four, with the same service names and price ranges as the business’s own pages. One page rewritten so the price question is answered in the first two sentences instead of in a table three screens down.

What moved. General recommendation shifted four points — reported as no change, because four points is drift. Price answers went from zero to being named in roughly a third of them.

What was not claimed. That the listings caused it. The report said what changed, when, and what moved after, and noted that the movement was concentrated in the question group the work had targeted.

How do you know whether it worked?

You need a baseline first, otherwise the next number has nothing to be compared to. Then:

1. Same questions, same phrasing, same number of repeats.
2. Reported per assistant, never averaged.
3. Raw answers kept, so any figure can be traced to the text behind it.
4. Movement under five points labelled as no change.
5. Attention to where the movement landed — concentrated in the targeted question group is weak evidence of effect; spread evenly is more likely drift.

Frequently asked

Can I submit my business to ChatGPT? No. There is no submission route.

Does having a ChatGPT plugin or GPT help? Not for being named in ordinary recommendation answers. Different mechanism.

Does blocking GPTBot hurt me? If you block it, your pages cannot be read when ChatGPT browses. Whether that costs you a mention depends on whether your own pages were being read in the first place — frequently they are not, but blocking removes even the chance.

How often does the answer change? Enough that a single check is a snapshot. Ask the same question five times and you will see the range for yourself.

Is it worth it for my category? Ask your buying questions and see whether ChatGPT names anyone at all. If it declines to recommend in your category, there is nothing to win this year.

A two-week plan

Days 1–2: find out where you stand

1. Write ten buying questions the way a customer would ask them.
2. Ask each five times. Record the answers verbatim, not as a summary.
3. Count the share that named you, by question type.
4. List every business named instead.

Days 3–5: find out what decided it

5. Ask the same questions in Perplexity, where sources are always shown, and in Google with AI Overviews. Record every cited domain.
6. Treat that list as the approximate source list for ChatGPT too — it will not be identical, and it will be close enough to act on.
7. Check whether you appear on each cited page, and whether what it says matches your own site.

Days 6–10: fix what you found

8. Get listed where you are absent, starting with the most-cited sources.
9. Correct what is wrong, especially services and hours.
10. Add prices where you publish them — price questions are where most businesses are missing.
11. Rewrite one page so the buying question you lose most is answered in its first two sentences.

Week four and after

12. Measure again, same questions, same repeats.
13. Report per assistant. Label anything under five points as no change.
14. If nothing moved, check whether the corrected listings have actually been re-read — often they have not yet.

What a realistic first result looks like

Not a transformation. In the measurement described above, the general recommendation share moved four points, which we reported as no change because that is what four points is.

The real result was elsewhere: a question type that had been at zero was no longer at zero. That is the shape of an honest first outcome — narrow, specific to the work done, and unimpressive to anyone hoping for a graph that goes up and to the right.

If someone shows you a first-month report where every number rose evenly, ask how many times each question was asked. Evenly rising numbers across a whole question set, in four weeks, is what drift looks like when nobody is measuring carefully.

Can an agency get me named faster than I can? Only by doing the same work with more hands. There is no privileged channel, and a supplier implying one has is telling you something that is not true.

Does having more reviews help? Probably, indirectly: reviews are part of what the sources an assistant reads are made of. It is not a lever you can pull quickly or honestly.

Where to start

Ask your own questions by hand today. Then, if the picture is worth measuring properly, the free check asks the questions for your business across four assistants and emails you what came back, with no call attached.

The method explains how the question set is written and why each question is asked several times. The full guide covers the work itself, and how to improve AI visibility is the step-by-step version.

One thing to do tomorrow

Ask ChatGPT your three most valuable buying questions, five times each, and write down every business named. If you are absent, you now know it; if you are present, you now know what your share roughly is. Either way you have a starting point that cost you twenty minutes and nothing else.