RANKING INTELLIGENCE · GPT-5-6

How to get your business recommended by ChatGPT

We record what ChatGPT actually does before it answers - every search it runs and every page it reads - and publish what that says about getting recommended. Every number here comes from those recordings.

4.4%
of recommended businesses got there because of a page someone published. That is the share you can influence.
53.7%
of everything ChatGPT read was a company's own website, not a review or a listing.
3.4
searches run, on average, before it answers a single question.
9
recorded answers behind these figures
Start here

Four ways a business ends up in the answer

We traced every recommended business back to whatever first put it there. Only one of these routes is open to a company that is not already famous.

Already known to the model
20.9%
Pulled from local business records
36.3%
Put there by a page it read
4.4%
No source we could identify
38.5%
What it searches for

The words it looks for, and what to do about each

ChatGPT writes its own searches before it answers a question. Every row below is a signal it put in those searches. Open one to see what it means for your site.

What ChatGPT looks for before it recommends anyone

Share of 26 searches it ran
9 recorded answers
Checking how current you are
19.2%

19.2% of searches carried a year, and some limited results to the last 30 days.

What to do about itDate your pages and keep the current year in the title where it is honest to do so. Undated pages lose the freshness check silently.
Looking for prices
15.4%

15.4% of searches asked about price, fees or plans.

What to do about itPut real numbers on the page. A quote form is not an answer to this search.
Reading one site in particular
15.4%

15.4% of searches were restricted to a single website.

What to do about itMost of these go to a company's own domain. Your site is the thing being read, so it has to answer the question directly.
Looking for the official source
11.5%

In 11.5% of the searches it ran, ChatGPT added the word "official" - it was hunting for a company's own authoritative page.

What to do about itHave one page that plainly states who you are, what you do, where you operate and what you charge. That is the page this search is trying to find.
Looking for reviews
3.8%

3.8% of searches asked for reviews or ratings rather than price or location.

What to do about itPublish a reviews page with real detail - the job done, the location, the date. Volume of stars alone carries nothing the model can quote.
LLMJESUS.COM · RANKING INTELLIGENCETAP ANY ROW FOR WHAT TO DO ABOUT IT
What it reads

The pages that actually get pulled

"Best" and "top" pages31.6% of pages read
31.6% of the pages it read were framed as best or top.
Do thisBeing listed on someone else's "best of" page is worth more here than publishing your own.
Pages with a year in the title28.9% of pages read
28.9% of the pages it read carried a year in the title.
Do thisDate your commercial pages and refresh them, so the year in the title is true.
Comparison pages11.1% of pages read
11.1% of the pages it read compared named options.
Do thisPublish an honest comparison including your competitors. It is the format most likely to be read back.
Numbered lists10.0% of pages read
10.0% of the pages it read were numbered lists.
Do thisAim to be an entry in the numbered lists that already rank in your category.
Pricing pages7.9% of pages read
7.9% of the pages it read were about cost.
Do thisPublish indicative pricing. It is read directly and it is rarely available.
Review pages5.8% of pages read
5.8% of the pages it read were reviews.
Do thisThird-party reviews of you are worth more than your own claims about you.
Guides and how-tos3.7% of pages read
3.7% of the pages it read were guides.
Do thisA buyer's guide with real numbers gives the model something specific to quote.
Tested or expert-led pages2.6% of pages read
2.6% of the pages it read claimed testing or first-hand expertise.
Do thisShow the work. First-hand detail is what separates a quotable page from a summary of other pages.
Award and accreditation pages0.5% of pages read
0.5% of the pages it read carried award or accreditation language.
Do thisList credentials with the issuing body, and link to the issuer as proof.
What we conclude

Read straight off the data

Most retrieval lands on the brand's own site

53.7% of 268 retrieved results were pages owned by a brand the answer named. Placement work cannot reach that share of the pool.

measuredbased on 9 recorded answers

The model scoped searches into sites we can get onto

1 of 4 domain-scoped searches targeted a domain no named brand owns. Those are obtainable listings and should be treated as the priority placement targets.

measuredbased on 9 recorded answers

What the pages have to answer

Share of the 26 captured fan-out queries carrying each signal: year_qualified 19.2%, pricing 15.4%, site_scoped 15.4%, official_lookup 11.5%. A page that cannot satisfy the common ones fails the check silently.

measuredbased on 9 recorded answers

Fields present but empty on every record

is_claimed, popularity_score, provider, ranking_score, review_highlights, reviews — anything computed from these on an earlier build is no longer computable from the transport.

measuredbased on 9 recorded answers

How we measured this

We record the network traffic behind real ChatGPT answers - the streamed response and the stored conversation - rather than scraping the page or asking the model what it did. One question per fresh conversation, on a real account, under a fixed daily limit.

  • 9 recorded answers across 8 categories on gpt-5-6.
  • Findings marked indicative rest on too few recordings to be firm. We publish them as such rather than rounding them up.
  • Answers vary between accounts, locations and settings. These come from a single account and a declared location.
  • Behaviour changes between model versions. Anything here is true of gpt-5-6 and is re-measured rather than assumed.

What would make these numbers wrong

Measured effects we can see in our own data. We publish them because a figure you cannot challenge is not evidence.

  • Our own question wording changes the answer. Prompts containing the word “best” pulled 39.0% best-titled pages; prompts that never said it pulled 24.9%. Read any “best” figure as partly an echo of how the question was asked.
  • Every capture comes from one account on one connection. Personalisation, location and account history all move these answers; we hold them constant rather than sampling across them.
  • The day-by-day bars show when we recorded, not how behaviour moved: our capture days so far covered different categories and markets. They become a trend once the same panel has run twice.
  • Categories here were chosen by us, not sampled at random, so they describe commercial questions of the kind we picked rather than all questions.
LLM JESUS · RANKING INTELLIGENCE · 9 RECORDED ANSWERS