WHAT THINGS COST
How do you get your site cited by ChatGPT?
What actually changes whether an assistant quotes you, and what cannot be bought.
Short answer
By being the clearest available source for a specific question: plain statements of fact, the same facts repeated as structured data, and a machine-readable surface an assistant can fetch. A two-week sprint to do all of it costs $2,500. Nobody can guarantee a citation, and anyone selling one cannot deliver it.
Assistants read differently from crawlers
A search crawler is building an index of pages to rank. An assistant is looking for a passage to quote, and it prefers text where the claim comes first, the evidence sits under it, and the qualifier is attached rather than implied. Marketing prose does the opposite: it builds to a claim, hedges it, and leaves the number in an image.
That is why the work is mostly restructuring rather than writing. The facts on most sites are already there, arranged to persuade a human reading top to bottom rather than to be extracted by something reading for an answer.
Say the same thing twice, once as data
If your price is in a sentence, a model has to parse the sentence and hope. If it is also in structured data, the number is unambiguous and matches what the page says. Prices, delivery times, locations, services and questions all have a schema, and using it removes the guessing.
The rule that matters: the data must equal the page. Structured data that disagrees with the visible text is worse than none, and it is one of the few things that reliably gets a site distrusted.
Publish a surface built for machines
An llms.txt is a plain-text summary of what the site is and where the important pages are, in the place an assistant can find it. It is cheap, and it means a model does not have to reconstruct your business from a navigation menu.
Being fetchable matters as much. Content that only exists after JavaScript runs, pages behind interstitials, and addresses rewritten into redirects that answer 404 are all ways a site accidentally hides itself. We found the last one on our own site during an audit, which is a good argument for measuring rather than assuming.
How to know whether it worked
From the server log. It records which assistant crawler fetched which page and when, which no dashboard can estimate accurately. Traffic arriving from assistants is harder: mobile assistants strip the referrer, so a link tagged with a campaign parameter is the only reliable count, and without it the number comes out several times too low.
Measure before changing anything, or the after number has nothing to sit next to.
What cannot be bought
Placement. There is no paid slot in an AI answer, and no relationship with a model provider that puts you in one. Anybody offering that is selling access that does not exist.
What can be bought is the removal of every technical reason not to cite you, and an honest measurement of what changed afterwards.
What moves the number
- Helps most
- Plain answer-first text, structured data that matches it, a fetchable site
- Helps less than people think
- Volume of content, keyword density, social signals
- Cannot be bought
- Placement in an AI answer. It does not exist as a product
The packages behind these numbers
Each one lists its full scope, what it excludes and how the work runs.