ChatGPT does not sit in a room and pick who deserves a citation. It retrieves. A person asks a question. The model looks for pages that already contain a named person, a specific fact, a date, and a URL on a domain it has reason to trust. If that page does not exist, the name is not in the answer. If the page exists but the sentence falls apart the moment it is lifted, the name is not in the answer either.

That is the whole job. Not fame. Not a journalist assignment. A dated public record that can survive being quoted out of context. Nobody at OpenAI is reading founder bios. Perplexity is not a talent desk. Gemini is not a publicist. Each is a retrieval system with a language model sitting on top. They quote what they can defend. They skip what they cannot. Silence is the default.

ChatGPT does not “decide” the way people mean

When someone says they want ChatGPT to cite them, they usually mean one of three things: the name appears in an answer, the answer links to a page about that person, or the answer treats that person as the source of a fact. Those are different outcomes. None of them is a promise. All of them start the same way: the model has to find a page.

Training data is stale, incomplete, and not a news wire. For a living person who is not already a household name, the useful path is search. The assistant looks at the open web, ranks what it finds, and copies a sentence it can stand behind. That sentence has to survive the lift. “She opened a practice in Austin” is garbage once it leaves the paragraph. “Jordan Hale opened a fractional CFO practice in Austin on 12 March 2026” is a fact. Put that fact on a high-authority publisher URL and the assistant has something to point at.

A citation is not a compliment. It is a copy-paste with a source. The source is a URL. The copy-paste is a sentence. If either half is missing, the name does not travel. A celebrity lives in the model weights. A consultant, a founder, a specialist, an operator lives on pages. If those pages are a personal site, a Medium draft, and a locked Instagram grid, retrieval has nothing high-authority to grab.

The model will often paraphrase. That does not relax the rule. A paraphrase still needs a subject. If the source sentence never names the person, the assistant will attach the fact to a company, a category, or nobody. Write as if the only thing that survives is one line. That is frequently what survives.

The four things retrieval actually needs

Write the sentence as if it will be the only thing the model keeps. That is not a metaphor. That is the failure mode.

A named person, a specific fact, a date, and a high-authority URL. Remove any one of those and the lift fails.

Four pieces have to be in the same breath:

  1. A named person. Full name. Not “the founder.” Not “the company.” Not “we.”
  2. A specific fact. A number, a place, a product, a hire, a launch. Something that can be wrong, which is how you know it is a fact.
  3. A date. A dateline, a year, a day. Undated copy looks like a brochure.
  4. A high-authority URL. The sentence has to live on a domain the assistant already treats as a source, not on a site the model has no reason to trust.

A named person with no fact is a bio. A fact with no name is an industry blurb. A dated claim on a blog nobody indexes is a diary entry. A USA Today URL full of “he” and “the firm” is a page the model can see and still cannot quote as about anyone in particular.

Marketing adjectives are worse than useless. “Visionary.” “Leading.” “Game-changing.” They are not facts, they trip wire editors, and they give an assistant nothing it can defend. A model that cites “the leading practice in the space” is guessing. A model that cites “Priya Shah opened a 12-person employment-law practice in Chicago on 4 June 2026” is repeating a record. One of those sentences can be checked. The other is a mood.

The URL is the part people try to fake with logo packs. A badge that says “as seen in” is not a citation. The live page is. If the URL 404s, it does not count. High-authority here is not a mystery: national newspapers, wire networks, finance publishers that already appear in search. That is the point of a wire, and the point of paying for the mastheads people actually screenshot.

A longer treatment of how to actually write those sentences lives in the companion piece on writing a press release AI will quote. This page is about why that shape is the one retrieval keeps, and why the domain the sentence sits on is half the citation.

Pronouns die when the sentence is lifted

Press copy is written in third person for a reason that has nothing to do with sounding formal. The reason is extraction. A language model does not quote a paragraph. It quotes a line. The line has to contain its own subject. “He closed a $2 million round” is a riddle. Who is he. When. According to whom. “He” died the moment the sentence left the previous one.

The same is true of “the company,” “the team,” “the platform,” and every other stand-in. They work on a homepage because the name is in the logo. They fail in a retrieval snippet because the logo is not in the snippet. They fail in a Perplexity source card for the same reason. They fail in a Gemini grounding snippet because Google’s snippet is also a lift.

Third person, full name, in the lead, and again in any sentence that might travel. That is the rule. It is also why a draft should not invent a quote. A fake quote is a fake fact. A real quote still needs the name in the attribution, not “said the CEO.” If the person did not supply a quote, the release ships without one. Thin and true beats thick and invented, especially on a masthead a stranger already believes.

Read a release out loud and delete every sentence that cannot stand in a search snippet. What remains should still name the person. If it does not, the release is a brochure with a dateline glued on. First person is a LinkedIn post. The wire will take third person or it will take nothing.

A sentence that survives, and one that does not

If the news is a hire, name the person hired, the role, the organisation, and the date. If the news is a book, name the author, the title, and the publication date. If the news is a practice, name the person who opened it, what it does, where, and when. Biography waits at the bottom. Retrieval reads the top.

ChatGPT, Perplexity and Gemini look at the same kind of page

Do not trust a screenshot of a chat. Models change, browsing gets turned on and off, and a lucky answer on a Tuesday is not a ranking. Anyone selling a walkthrough of “this is exactly how ChatGPT cites you” is guessing or out of date. The high-level picture is enough, and it is the same picture for all three.

ChatGPT, when it can search, pulls live pages and writes an answer from them. It prefers sources it can name. A news URL with a dateline beats a personal blog. A named, dated claim beats a vibe. It will not reliably surface a person who exists only on social, and it will not keep citing that person next month because one chat went well this month.

Perplexity is built as a citation machine. It shows sources as a feature, not a footnote it might skip. That makes the URL even more of the product. A page on a recognised publisher is a source Perplexity can display. A Carrd site is not. It still needs the same four pieces. It does not invent a source to be polite.

Gemini sits on Google. Grounding, when it happens, looks a lot like search: news, knowledge panels, pages the index already trusts. A Google News result for a name is not a guarantee Gemini will speak that name. It is the closest thing to a shared substrate. People who want the search version of this document should read how to get into Google News with a press release. The page you want in Google News is the same shape of page an assistant can cite: timestamped, named, on a domain the index already treats as news.

The interfaces move. The retrieval problem does not. Named person, specific fact, date, high-authority URL. That is the input. The output is sometimes a citation, sometimes a paraphrase with a link, sometimes silence. Buying a release does not purchase the output. It purchases the input.

A press release is syndication, not a reporter

Agencies sell the dream of a journalist. A wire sells a reprint. Cited sends the approved release through a third-party syndication partner. A release goes onto a feed. Publisher sites reprint the copy. Some of those sites are ones a stranger would recognise — AP News, USA Today, Yahoo Finance. Most of the rest are not. Pickup is not coverage. Coverage is a reporter who chose the story, called someone, and wrote their own sentences. Syndication is the original copy, dated, sitting on someone else’s domain.

That distinction is not a downgrade. It is the point. A reporter might never call. The record still exists. Assistants do not need a reporter. They need a URL they can defend. The person who waits for coverage before they exist on the record is waiting in the wrong queue.

People pay for USA Today and AP because those domains are the ones a stranger already believes, and the ones a screenshot actually looks like. On Cited, USA Today is $179 and AP News is $199 as add-ons on Get known, and both are included in Famous. The customer gets the live URLs. If the URL is not there, it does not count. A USA Today press release is a live page, not a newsroom assignment.

The wire also has hard limits. The body has to be 250–1,250 words. That is not a style suggestion. Under 250, the wire rejects it. Over 1,250, the same. Padding a 180-word announcement with adjectives to clear the floor is how people produce copy no assistant will quote. If the facts are short, add context the facts already imply. Do not add a vision statement.

What Cited actually puts on the record

Cited is one job: a press release about a person, written so a lifted sentence still contains the name, then put on the wire. The draft is AI. The facts are not. The person types the news — a launch, a hire, a book, a practice. The model writes third person, inverted pyramid, inside the word limits. It does not invent a quote, a customer, or a number. If the facts are thin, the release is thin. That is safer than a confident lie on USA Today. Every word is edited by the person it is about before it ships. How Cited works is the short version of that pipeline.

Distribution reaches 500+ outlets on the packages that include that reach. Orders do not spend automatically. They sit until the release is actually ready. Prices are pay-once, no subscription:

A personal press release is the right shape. Company PR is for the brand. Retrieval on a person’s name needs the person in the lead. None of this is a guarantee that ChatGPT, Perplexity, or Gemini will cite anyone. Models retrieve what they retrieve. A well-formed record on a high-authority domain is the input those systems are built to use.

What still will not get a name cited

A job title is not news. “Founder and CEO” is a LinkedIn headline. The wire needs a thing that happened. A personal site is not a news domain. A logo pack is not a URL. A first-person announcement is not extractable. “We are excited” dies on contact with a snippet. A guarantee is a lie. Anyone selling “we will get you into ChatGPT” is selling a lottery ticket with a slogan on it. The honest product is a dated page. The citation, if it happens, is a consequence of retrieval, not a line item.

Put the sentence on the record

Write the news. Name the person in every sentence that might travel. Date it. Put it on a domain an assistant already trusts. That is the work. Everything else — screenshots of chats, logo packs, threads about “GEO” — is commentary on work someone else did, or did not do.

Cited will draft it. The person it is about will edit it. Cited will syndicate it. The next step is the compose page, not another essay about how models might one day work.

Write the release.

Questions

Will ChatGPT cite someone after one press release?

No one can promise that. ChatGPT, Perplexity and Gemini retrieve pages; they do not assign citations. A named person, a specific fact, a date and a high-authority URL is the input those systems can use. Silence is still the default.

Why does a press release have to be in third person?

Pronouns die when a sentence is lifted. “He closed a round” is a riddle out of context. Third person with the full name in the lead, and again in any sentence that might travel, is how a quote still has a subject.

Is a USA Today or AP News placement a reporter writing the story?

No. Distribution is handled through publisher syndication. Pickup is syndication: the original copy, dated, reprinted on publisher sites. Coverage is a reporter who chose the story. On Cited, USA Today is $179 and AP News is $199 as add-ons on Get known, and both are included in Famous. The customer gets the live URLs.

How do ChatGPT, Perplexity and Gemini differ for this?

At a high level they do not. ChatGPT can search live pages. Perplexity shows sources as a feature. Gemini sits on Google’s index. Interfaces move. The page they can cite does not: a timestamped, named fact on a domain they already treat as a source.

What does Cited actually write and ship?

An AI draft from user-supplied facts only, edited by the person it is about, 250–1,250 words, then sent through publisher syndication. Indexed is $79, Get known is $149, Famous is $399. Start on the compose page at /write.