Lumiono

How to get cited by ChatGPT: the seven signals that decide it

ChatGPT does not cite a domain because it ranks on Google. It cites pages it can fetch, quote and corroborate across independent sources. Below are the seven signals that decide it, the checks we run on every audit, and the parts that are outside anyone's control.

First, three different things get called “cited”

Named — the model writes your business name in the answer. Linked — the answer shows your URL as a source chip (ChatGPT Search, Perplexity, AI Overviews). Quoted — the answer repeats a sentence from your page. They are driven by the same inputs but show up at different speeds: Perplexity and AI Overviews attach links within weeks; a plain ChatGPT answer usually names brands without showing a source, which is why “I have no backlink from ChatGPT” is a misunderstanding, not a bug.

Signal 1 — the crawler can actually read the page

Open yourdomain.com/robots.txt. If GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot or Google-Extended are blocked, the assistant only knows you from second-hand data — usually a directory listing with a stale phone number. Second check: the content must exist in the HTML. A page that renders its services only through JavaScript is a blank page to most crawlers. Test with curl -s yourdomain.com/service and look for the actual sentences.

Signal 2 — one question, answered in the first two sentences

Models retrieve passages, not pages. A heading written as the buyer's question (“How much does an emergency plumber cost in Austin?”) followed immediately by a number is retrievable. The same page with the answer three paragraphs below a brand story is not. This is the cheapest fix on the list and the one most sites skip.

Signal 3 — the facts are identical everywhere

Name, address, phone, service area, hours, price band: these must match character-for-character on your site, your Google Business Profile, your Apple Maps listing, the directories and any press mention. Conflicting addresses are the most common reason a model refuses to name a business it can clearly see. Pick one canonical form and overwrite every other copy.

Signal 4 — a machine-readable entity exists

JSON-LD on the homepage — LocalBusiness or ProfessionalService, plus Service, Offer and FAQPage, with sameAs links to every profile you control. Two rules: every value must match visible text, and one entity per site. A schema that contradicts the page is worse than no schema, because it makes the record ambiguous.

Signal 5 — independent pages repeat the fact

This is the part no on-site edit can replace. When four or five pages you do not own state the same thing — a chamber listing, a trade directory, a “best X in Y” roundup, a real review profile — the model treats it as reliable and names you. A brand with a perfect site and zero third-party mentions loses to a mediocre site named in three lists. It is measurable: search your category and count how many results are roundups, then count how many of those name you.

Signal 6 — the page looks maintained

Dated statements (“prices as of October 2026”), a real lastmod in the sitemap, and prices that match today beat an undated page for time-sensitive questions. Stale prices are worse than missing ones: an assistant that quotes your 2023 price makes you look wrong on the call.

Signal 7 — your headings match how buyers ask

Buyers type “who does emergency AC repair in Mesa on Sunday”, not “our services”. Every service page should carry the ten to twenty-five real phrasings — city, urgency, price, qualification — as headings with an answer under each. This is also the structure that wins Google AI Overviews.

The 60-minute self-audit

CheckWhat good looks like
robots.txtGPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended all allowed
Raw HTMLcurl shows the service sentences without JavaScript
llms.txtFacts in plain sentences: what, who, where, price band, hours, contact
SchemaOne LocalBusiness/Organization block, values match the page, sameAs filled
NAPIdentical on site, Google Business Profile, Apple Maps, top 3 directories
SitemapReal lastmod dates, no 404s, submitted in Search Console and Bing
Answer pagesQuestion-shaped H2s, answer in the first two sentences
Third-partyAt least 3 independent pages stating the same facts

What does not work

How long it actually takes

Baseline measurement takes three days: twenty buyer questions against two or three models, one clean run each, counting how many answers name the brand. After the fixes ship: Perplexity and Google AI Overviews typically start attaching links in two to four weeks; ChatGPT Search naming a new business usually takes four to eight; the compounding effect on roundups and directories runs past ninety days. Anyone promising this week is guessing.

Questions

Does ChatGPT use my Google ranking?

Partly and indirectly. ChatGPT Search can surface pages that also rank, but the plain model answers from what it has read. A page ranking #1 with no third-party corroboration still loses to a smaller brand named across several independent pages.

Can I pay ChatGPT to mention my business?

No. There is no ad unit that inserts a brand into a model answer. Sponsored placements exist in some assistant surfaces for shopping, not for “who should I hire” answers.

Does adding llms.txt guarantee a citation?

No. It removes ambiguity about your facts and makes the page cheap to parse, which raises the odds the model gets your details right. It does not create the corroboration signal that decides whether you are named at all.

How do I check whether I am already cited?

Run twenty real buyer questions (category + city + urgency) through ChatGPT, Perplexity and Gemini with browsing on, save the answers, and count the runs that contain your brand name. Repeat monthly with the same questions. That number is your visibility score.

What is the fastest single fix?

Making the answer the second sentence under a question-shaped heading on your five highest-value service pages. It is on-site, it takes an afternoon, and it is what gets retrieved.

Related

The 12-point AI visibility audit · How the process works · What it costs · Our own baseline: 0 of 40


Get the free AI-readiness audit

Send your domain and service area — you get the exact scope, price and delivery date in writing. Launch is $1,500 one-time, tracking is $500/month.

Run the free audit See pricing