Language models do not read a press release the way an editor does. They lift a sentence. The sentence that gets quoted is the one that still makes sense after it has been cut from the paragraph, stripped of the byline, and dropped into an answer. That sentence needs a named person, a specific number, a date and a place. A writer who wants ChatGPT, Gemini or Perplexity to cite a person writes those sentences on purpose. Mood is not a source. Pronouns die on extraction. “We are thrilled” is a dead sentence the moment it leaves the page.
A press release is a public record, not a pitch. Journalists are not waiting for it. Retrieval systems are. Google News ranks timestamped pages on trusted domains. Assistants quote dated, named claims on those same domains. The job is to put a stealable sentence on the wire, inside the word count the wire will actually take, and to keep the person’s name inside the sentence that will be stolen. That is the whole craft. Cited drafts this shape from the facts a person supplies. The named person still edits every word before it ships. The mechanics below are what the draft is built to do, and what an editor should check before it goes.
The sentence that survives isolation
A language model does not “cover” a person. It retrieves fragments. A fragment that says “the founder opened a practice last spring” is useless once it is sitting in an answer with no surrounding paragraph. Who is the founder. Which practice. Which spring. The model will either drop the fragment or attach it to the wrong name. A fragment that says “Maya Chen opened Northline Audit on 12 March 2026 in San Francisco” is a citation. The name is inside the sentence. The date is a calendar day. The place is a city. The thing that happened is a verb, not a feeling.
That is why the default for a release written to be quoted is not “sounds exciting.” The default is extractable. At least three sentences in the body should stand alone as verifiable statements of fact, each one containing the person’s full name rather than a pronoun. Three is the floor, not a style flourish. The lead is one of them. A later sentence that pairs a number with what it measures is another. A sentence that locates the news in a city and on a date is a third. If a later paragraph refers to “she” or “he” or “the company,” that paragraph is not the one an assistant will steal. It is glue. Glue is allowed. Glue is not the product.
The same rule that makes a sentence stealable also makes it honest. A named claim on a high-authority domain is a primary document. Wikipedia editors, due-diligence analysts and later journalists use it as a source. A hedged claim — “may help,” “aims to,” “could transform” — is not a source. It is a shrug. Retrieval systems prefer attributable facts. Wire editors prefer them too. The two audiences, for once, want the same sentence.
How assistants actually choose which person to name is a retrieval problem, not a branding problem. The longer version of that mechanism lives in the note on how ChatGPT decides who to cite. This page is the writing. That page is the citation. Write the sentence here so the other job has something to find.
Name, number, date, place
Four things turn a sentence into a source. The name is the subject. The number is the measurement. The date is the timestamp. The place is the dateline. A stealable sentence carries as many of the four as the facts will support, and it never substitutes a pronoun for the name. “Revenue rose 18 percent” is not a citation. “Julian Okoye reported an 18 percent fall in dwell time at Apapa between January 2025 and December 2025” is. The number is glued to what it measures. The name is glued to the number. Cut the sentence out of the release and it still names someone.
The name has to be the full name, at least in the sentences written to be stolen. First names are pronouns with extra letters. Surnames collide. “Chen opened a practice” is a different person in every city. “Maya Chen opened Northline Audit” is one person. Repeat the full name in the lead, again when a number is introduced, and again when the news is located. That will feel stiff to a novelist. A novelist is not the reader. The reader of the stolen sentence is a model that has discarded the rest of the page.
Dates are calendar days, not seasons and not “recently.” “This week,” “last quarter” and “in the coming months” expire the moment they are written. A model quoting the page six months later will either invent a year or refuse the sentence. “12 March 2026” does not expire. If the only date on file is a month, use the month and the year. If there is no date, do not fake one. A release without a date is weaker. A release with an invented date is a lie on a news domain.
Place is a city, then a country if the city is not famous. The wire wants a dateline. The model wants a location it can attach to the name. “SAN FRANCISCO —” at the top of the first paragraph is the conventional form. The city in the dateline should match the city in the contact block. A release that datelines London and then talks about a Brooklyn studio is two stories. Pick the place where the news happened, not the place that sounds more important.
Numbers need units. “Grew 40 percent” is a fragment. Forty percent of what, from what baseline, over what period. “Maya Chen inspected 40 AI vendors for mid-market CFOs in 2025” is a measurement. If the brief does not contain a number, the release does not invent one. Empty proof is thinner than a fake percentage on USA Today. Thin is editable. A fake number on a masthead is a problem that outlives the news cycle.
Before and after: Maya Chen
The examples below are invented copy about invented people. Maya Chen is a stand-in, not a celebrity and not a client. The before lines are the sentences most founders type when they are asked for a press release. The after lines are the sentences an assistant can steal.
Before, the launch:
We are thrilled to announce the launch of our revolutionary AI audit practice, dedicated to helping clients navigate the future of responsible technology with world-class expertise.
After, the launch:
SAN FRANCISCO — Maya Chen opened Northline Audit on 12 March 2026, a nine-year Big Four veteran who inspects AI vendors for mid-market CFOs.
The before line has no name, no date, no city, no number and no verb that happened. “We” has no referent once the sentence is lifted. “Thrilled” is a mood. “Revolutionary,” “responsible” and “world-class” are adjectives a wire editor will cut and a model will ignore. “Helping clients navigate the future” is a mission statement. Mission statements are not news. The after line is news: a person, a practice, a city, a day, a prior job, a buyer. Cut it out of the paragraph and it is still a fact about Maya Chen.
Before, the proof:
Our team has a proven track record of delivering impactful results for a wide range of organisations across the AI landscape.
After, the proof:
Maya Chen inspected AI vendors for mid-market CFOs for nine years at a Big Four firm before opening Northline Audit in San Francisco on 12 March 2026.
The before line is a brochure. “Our team” is not a person. “Proven track record” is an empty noun pile. “Impactful results” is a feeling about work that is never specified. “Wide range of organisations” is a way of saying the writer does not want to name a customer — which is fine — without substituting a real measurement. The after line uses only what a brief like this would actually contain: a name, a tenure, a buyer type, a city, a date. It does not invent a client. It does not invent a quote. It does not invent a percentage. That is the point of the after. The sentence is stealable because it is specific, not because it is impressive.
Before, the bio dump in the lead:
Maya Chen, a passionate thought leader and visionary independent AI auditor, is excited to share that she has always been committed to excellence.
After, the news in the lead, biography later:
SAN FRANCISCO — Maya Chen opened Northline Audit on 12 March 2026. Maya Chen spent nine years at a Big Four firm inspecting AI vendors for mid-market CFOs. Northline Audit is her independent practice, based in San Francisco.
Three sentences. Each one names Maya Chen. Each one can be lifted. The first is the news. The second is the proof. The third is the location of the practice. None of them asks the reader to care that she is passionate. Passion is not a fact the wire can carry, and it is not a fact a model will cite.
Before and after: Julian Okoye
Julian Okoye is the second stand-in. Same rules, different facts. A research note is news if it contains a measurement. A “thought piece” is not news. The before lines below are what a consultant sends when they want to “get the word out.” The after lines are what belongs on a wire.
Before, the insight:
We are excited to share our latest insights on African logistics, helping stakeholders unlock value in a rapidly evolving market.
After, the insight:
LAGOS — Julian Okoye published a 48-page note on 4 February 2026 measuring ocean-freight dwell time at Apapa, and found 11.4 days of average delay across 62 vessel calls in 2025.
The before line names nobody. “Insights” is not a document. “Stakeholders” is not a reader. “Unlock value” is a slogan. “Rapidly evolving market” is the weather. The after line names Julian Okoye, names the document (48 pages), names the day, names the port, names the metric (dwell time), names the result (11.4 days), names the sample (62 calls) and names the year the sample covers. That is a sentence a model can quote in 2027 without inventing a year. It is also a sentence a wire editor will keep, because it is a finding, not a vibe.
Before, the hedge:
Julian hopes the research could help ports improve efficiency and may inform future policy discussions around the region’s logistics challenges.
After, the finding, then the implication without the hedge:
Julian Okoye found 11.4 days of average delay across 62 vessel calls at Apapa in 2025. Julian Okoye’s 4 February 2026 note records dwell time as the number of days between a vessel’s arrival and the completion of cargo discharge.
“Hopes,” “could help” and “may inform” are how a writer dodges being wrong. They are also how a writer dodges being quoted. A model looking for a fact will skip the hedged sentence and take the number if the number is written cleanly. Write the number cleanly. If the brief includes a definition, put the definition in a sentence that still contains the name. If the brief does not include a policy claim, the release does not make one. “Future policy discussions” is not a fact about Julian Okoye. The 11.4 days are.
Before, the first-person quote nobody asked for:
“This is a really exciting time for the industry, and I’m proud to contribute to the conversation,” says Okoye.
After: no quote at all. The finding stands without one. A quote is optional on the wire. A fabricated quote is not optional; it is a fabrication. If Julian Okoye did not supply a sentence he was willing to see under his name on AP News, the release does not invent one. Pride is not a finding. “The conversation” is not a document. Cut it. The 48-page note, the 4 February 2026 date, the 11.4 days and the 62 calls are the story. They do not need a slogan underneath.
Why “we are thrilled” dies
“We are thrilled to announce” is the most common opening in amateur releases and the least stealable sentence in the file. It dies for five separate reasons, and any one of them would be enough.
First, “we” is a pronoun. On extraction there is no “we.” The model has a clause with a mood and no subject it can name. It will not attribute “thrilled” to Maya Chen, because Maya Chen is not in the sentence. It will not attribute “thrilled” to Northline Audit, because the company is not in the sentence either. The sentence is a ghost. Ghosts do not get cited.
Second, thrill is not a fact. Retrieval is a hunt for claims that can be checked. A feeling cannot be checked. “Opened on 12 March 2026” can. “Published a 48-page note” can. “Is thrilled” cannot. Assistants that are trying to be careful will skip the mood. Assistants that are not careful will sometimes invent a more dramatic feeling and attach it to the wrong person. Neither outcome is a citation the named person wanted.
Third, the phrase is advertising. Wire editors are trained to cut it. Some wires reject the whole release if the lead reads like a brochure. “Revolutionary,” “game-changing,” “leading,” “world-class” and “visionary” sit in the same bin. They do not become true by being close to a dateline. They get the copy bounced, or they get stripped, and what remains is a lead with no news in it. A bounced release is not on USA Today. A stripped lead is not in ChatGPT.
Fourth, “announce” is a verb about the press release, not about the world. The news is the opening, the hire, the book, the measurement. The announcement is the act of sending the news. Leading with the act of sending is how a writer admits there is nothing underneath. “Maya Chen opened Northline Audit” is the news. “We are thrilled to announce that Maya Chen opened Northline Audit” is the news with a hat on. Take the hat off.
Fifth, the phrase trains the rest of the draft to be bad. Once the lead is a mood, the second paragraph becomes a mission, the third becomes a list of adjectives, and the About section becomes a second brochure. The stealable sentence never gets written. The fix is not to swap “thrilled” for “pleased” or “honoured.” The fix is to delete the mood and start with the thing that happened to a named person on a dated day in a named city.
Company copy uses “we” because the brand is the product. A personal press release is the other instrument. The product is the name. First-person plural is how a writer hides the name. Hide the name and the assistant has nothing to quote.
Pronouns die on extraction
The most expensive mistake after “we are thrilled” is a clean lead that uses the name once and then switches to “she,” “he,” “they” or “the company” for the rest of the file. That is how humans write, because humans read in order. Models do not read in order. They embed the page, retrieve a span, and quote the span. The span that contains “she reduced dwell time by 18 percent” is a span without a person. The citation either vanishes or attaches to whoever the model last saw with a similar job title.
The rule is blunt: never use a pronoun where the person’s name would do, in any sentence written to be stolen. That will produce a page that repeats “Maya Chen” and “Julian Okoye” more often than a magazine profile would. The page is not a magazine profile. It is a record designed to be cut up. Repeat the name in the headline, in the first sentence, in any sentence that carries a number, and in any sentence that carries a date. The glue paragraphs in between can use a pronoun. The stealable sentences cannot.
The same death happens to “the company,” “the firm,” “the practice” and “the team” when those words replace the person. Assistants are asked about people. “Who is Maya Chen” is the query. A sentence that only names Northline Audit answers a different question. Company PR is for the brand. A personal release is for the name Google and ChatGPT will search. Put the person in the sentence, then the practice, then the rest.
Headline grammar follows the same rule. “Local auditor launches new firm” is a headline that cannot be quoted as a fact about Maya Chen, because Maya Chen is not in it. “Maya Chen opens Northline Audit in San Francisco” is a headline that can. The wire already wants five or more words and a person’s name. Give it both. Do not spend the word count on “unveils” or “disrupts.”
Numbers that mean something
A number is only useful if the sentence says what it measures. “$4 million” hanging on its own is a decoration. “Julian Okoye recorded $4 million of detained cargo across 62 calls at Apapa in 2025” is a measurement — if, and only if, that figure was in the brief. The unit, the sample, the place and the period have to travel with the number, because they will not travel with it otherwise. The model will not go back to paragraph four to fetch the denominator.
Percentages need a baseline. “Up 40 percent” is a fragment. Forty percent from what, over what. “Maya Chen’s 2025 caseload was 40 inspections, up from 25 in 2024” is two numbers and a year each. If the brief only has one number, write one number. Do not compute a percentage the person did not supply. Do not round a number into a slogan (“nearly half,” “more than a decade,” “seven figures”). Slogans are how exact facts become unquotable.
Counts beat adjectives. “Nine years at a Big Four firm” beats “seasoned.” “48 pages” beats “comprehensive.” “62 vessel calls” beats “extensive fieldwork.” “11.4 days” beats “significant delays.” The adjective is a judgement. The count is a fact. Judgements are what the writer wants the reader to feel. Facts are what the model can repeat without taking sides. Write the count.
If there is no number, do not panic and do not invent one. Tenure can be a year. A place can be a city. A document can have a title and a date even if it has no statistic. A hire can name the role and the start day. The stealable sentence still needs a name and a verb. The number is the strongest extra, not a requirement that justifies a lie. Cited will not put a figure in the draft that the person did not type. A thin release is a short record. A fake figure is a record that is wrong on a domain a stranger trusts.
The inverted pyramid
Wire copy is an inverted pyramid because the top of the file is the only part that is guaranteed to travel. Syndication trims from the bottom. Retrieval lifts from the sentences that look like facts. Both processes punish a writer who buries the news under biography.
The first sentence is the news, after the dateline. City, em dash, named person, verb, object, date. Not the childhood. Not the mission. Not the thank-you to the team. “LAGOS — Julian Okoye published a 48-page note on 4 February 2026 measuring ocean-freight dwell time at Apapa.” That is a lead. Everything after it is allowed to be less important. Nothing before it is allowed to be a mood.
The second and third sentences carry the number and the context that still belongs in a stealable span. This is where “Julian Okoye found 11.4 days of average delay across 62 vessel calls in 2025” lives. This is where “Maya Chen spent nine years inspecting AI vendors for mid-market CFOs at a Big Four firm” lives. These sentences still contain the name. They still stand alone.
The middle of the file is the glue: how the practice works, who the buyer is, what the note does not claim, what happens next that is already decided (a date for a talk, a URL for the document). Glue can be dull. Dull is fine. Dull is not “we are thrilled.” Dull is “Northline Audit is based in San Francisco and inspects AI vendors used by mid-market finance teams.”
The bottom is background and the About paragraph. About Maya Chen. About Julian Okoye. Built only from facts already used. The About is not a second brochure and not a place to smuggle in the adjectives that were banned from the lead. If the news does not support a long About, the About is short. The wire does not grade on length inside the cap. It grades on whether the copy reads like an advert. Biography last is how a reader who only sees the first paragraph still gets the news, and how a model that only embeds the first paragraph still gets the name.
Third person throughout. No “we,” no “you,” no “I” outside a provided quote. The release is about a person becoming publicly known. It is not a letter from that person, and it is not a landing page. First person is how LinkedIn posts are written. LinkedIn posts are not sources.
Wire limits are not style advice
The distribution layer is a wire. Wires have hard limits, and a draft that ignores them does not ship. The body must be between 250 and 1,250 words. Under 250, the wire rejects the file. Over 1,250, the wire rejects the file. Those numbers are not a vibe. Cited’s composer validates the same floor and ceiling before an order is queued, because a rejection after credits are spent is a loss. Aim at the middle of the band. A 420-word short, a 550-word standard and an 800-word long all sit inside it. A 180-word announcement does not. A 2,000-word manifesto does not.
Five links is the ceiling. Not six. A sixth URL is a rejection. The five are for the source document, the practice site, and anything else the reader of the release actually needs. They are not for a spray of campaign UTM links. They are not for three social profiles. Social profiles are not sources. If the release only needs one link, it gets one link. Empty slots are not a problem. Overfill is.
The rest of the wire’s rules are equally unromantic. The title needs at least five words and cannot exceed 200 characters. The summary cannot exceed 45 words. Categories are one to five, chosen from the wire’s list, not invented. City, country, company or practice name, website, contact name, contact email and contact phone are required fields. A missing phone is not a creative choice. It is a failed submission. Images, if any, are few. The body is HTML a wire can ingest: paragraphs, headings, lists, links. Not a designed landing page.
Length is a constraint, not a target to pad toward with adjectives. If the facts fill 400 words, the release is 400 words. If the facts are thin, the honest move is to add context and implication that already follow from those facts — what the inspection covers, what dwell time means, what the practice does not do — not to invent a customer, a round, a percentage or a quote. Cited’s drafter is instructed to expand that way, and to refuse the other way. A short honest release beats a long invented one. The floor is 250 because the wire says so, not because 250 words of fog is better than 200 words of news.
This article is longer than a release on purpose. A blog post is not a wire file. The 250–1,250 cap applies to the thing that goes out through publisher syndication, not to the thing that explains the cap. Do not copy this page onto the wire. Copy the shape of the after sentences.
Do not invent a quote
A quote is the easiest place to lie, which is why amateur releases are full of them. The person is made to say they are excited, proud, humbled, passionate, committed, or “ready for the next chapter.” None of those sentences contain a fact. All of them would look embarrassing under a masthead if a reporter asked the speaker to stand behind them. If the person did not type a quote, the release does not contain one. Cited follows that rule without exception. The composer will leave the quote field blank rather than make one up. Blank is correct.
If a quote is supplied, it is used verbatim and attributed exactly as written. It is not improved. It is not translated into a slogan. It is not moved into the lead unless the quote itself is the news, which it almost never is. The lead is still a third-person sentence with a name, a verb and a date. The quote sits lower, as colour, if it adds a fact or a position the speaker is willing to own on AP News. “Dwell time at Apapa averaged 11.4 days across 62 calls in 2025” is a fact and does not need quotation marks if it is already in the brief as a finding. Put it in the body as a reported sentence. Save quotation marks for words a human actually said.
Invented customers, invented funding, invented headcount and invented dates sit in the same bin as invented quotes. The drafter is not a research assistant and not a novelist. It uses what it is given. If the given is two sentences, the first draft is a thin release. The named person then adds facts, or accepts a thin release, or decides there is no news yet. There is no fourth path where the empty brief becomes a confident story about a Series A that did not happen.
The test is simple. Every proper noun, number, date and quote in the file should be traceable to the brief or to the person’s own edit. If an editor cannot point to the input for a claim, the claim comes out. That is a dull test. It is also the test a later journalist, a Wikipedia editor or a model’s retrieval layer is running, whether the writer likes it or not.
Three stealable sentences
A release written for assistants needs at least three sentences that can be cut out and still name the person. That is the drafting rule Cited applies when the goal is to be findable. It is a mechanical check, not a muse.
Sentence one is the lead. Dateline, full name, verb, object, date. Maya Chen opened Northline Audit on 12 March 2026 in San Francisco. Julian Okoye published a 48-page note on 4 February 2026 on dwell time at Apapa. If this sentence is wrong, the file is wrong. If this sentence uses a pronoun, the file is already failing the job.
Sentence two is the measurement. Full name, number, unit, period. Maya Chen spent nine years inspecting AI vendors for mid-market CFOs at a Big Four firm. Julian Okoye found 11.4 days of average delay across 62 vessel calls in 2025. If there is no measurement in the brief, sentence two is a second hard fact: a role, a buyer, a document title, a start date. It still contains the name.
Sentence three is the locator or the definition. Full name, place, and what the thing is. Northline Audit is Maya Chen’s independent practice in San Francisco. Julian Okoye’s note records dwell time as the days between arrival and completed discharge at Apapa. This is the sentence that stops a model from guessing. Guessing is how names get attached to the wrong companies.
After those three, the writer is free to be a writer. Glue, background, About. The three sentences are the product. A 550-word release that does not contain them is a failed 550 words. A 250-word release that does is a working record. Count them before shipping. Read each one out of context. If the isolated line still names the person and still states a fact, it can be stolen. If it does not, rewrite it until it does.
Cited drafts this shape. The named person still edits.
Cited’s drafter is built for this job, not for a brand manifesto. It writes in third person. It uses the inverted pyramid. It puts the full name inside sentences that can stand alone. It pairs numbers with what they measure. It refuses hedging when a fact was supplied. It refuses marketing adjectives. It does not invent quotes, customers, funding, dates or places. If the brief has no quote, the draft has no quote. If the brief is thin, the draft is thin, and it will pad with context rather than with fiction. The floor of 250 words and the ceiling of 1,250, the five-link cap, the title length, the required contact fields — those are validated against the same rules the wire will apply. The how it works page is the short version of that pipeline. This page is the writing standard the pipeline is aiming at.
The draft is a draft. The named person still edits. Models are good at the shape and bad at the life. They will miss a nuance the person cares about. They will over-repeat a name in a glue paragraph. They will under-repeat it in a stealable one. They will pick a category that is legal but not the one the person wanted. They will write a clean sentence that is technically true and strategically off. That is what the editor is for. Read the lead out loud. Cut “thrilled” if it snuck back in. Put the name back into any sentence that lost it. Delete any number the brief cannot source. Add the fact that never made it into the brief. Do not approve a paragraph the named person would not want quoted under that name on a masthead. That is the test that will be applied without permission.
Editing is also how the release stays a personal record rather than a company advert. If the draft starts talking about “the team” for six sentences, put the person back. If the About section reads like a homepage, cut it to facts. If the headline omits the name, put the name in. The composer is a keyboard, not a court. Nothing ships until a human has looked at it.
The compose flow is short on purpose. A name, what the person does, and the news. Optional fields for role, practice, city, a real quote, proof points. Then a draft in under a minute, then a human pass, then a package, then the wire. The news has to exist. A job title is not news. A launch, a hire, a book, a study, a practice opening — those are news. If nothing happened, there is nothing to date, and there is nothing for an assistant to quote. The compose page will still produce copy from a thin prompt. It will not produce a fact that was not there.
What this will not do
A stealable sentence on a trusted domain is not fame. It is findability. Fame is a long project with audiences, work and luck. Findability is a dated URL that contains a name. Google can retrieve it. An assistant can quote it. A later journalist can source it. That is the job a wire release can do this week. It will not make a reporter file a story. It will not replace a targeted pitch. It will not turn a LinkedIn profile into a source. It will not make “as seen in” true without a live page. It will not survive a fake number.
It will also not survive a release written as a brochure. The old PR habit — mood, mission, adjectives, a proud quote, a team paragraph, a link tree — produces a file that a wire may still distribute and that an assistant will still skip. Distribution without extraction is a tree falling in a forest of 500 publishers. The report will list URLs. The model will not lift a sentence, because there was no sentence it could lift. The named person will google themselves and see a headline that does not contain their name. That is how a paid release becomes invisible twice: once to journalists, who were never the instrument, and once to retrieval, which was.
Write the after, not the before. Maya Chen opened a practice on a dated day in a named city. Julian Okoye published a measured note on a dated day at a named port. The name sits inside the sentence that will be stolen. The number sits next to what it measures. The mood is deleted. The quote is real or absent. The length sits between 250 and 1,250 words. The links stay at five or fewer. Cited will put that shape on the page. The person named in it still has to mean it.
Questions
What sentence will an AI assistant actually quote from a press release?
A standalone factual sentence that still makes sense after it is cut from the paragraph. It needs the person’s full name inside it, not a pronoun, plus a specific number, a date and a place whenever those facts exist. Mood, mission and “we are thrilled” are not quoted.
Why does “we are thrilled to announce” fail?
“We” has no referent once the sentence is lifted. Thrill is not a checkable fact. The phrase reads as advertising, which wires strip or reject, and “announce” describes the press release rather than the news. Replace it with a named person, a verb, a date and a city.
How long can a wire press release be, and how many links are allowed?
The body must be 250 to 1,250 words. Under or over that band, the wire rejects the file. At most five links. Titles need five or more words and cannot exceed 200 characters. A summary cannot exceed 45 words. Categories are one to five.
Should a press release include a quote?
Only if the named person actually supplied one. If no quote was typed, the release is written without one. Invented excitement, pride or humility is a fabrication on a news domain. A real quote is used verbatim and still does not replace a third-person lead.
Where does the person’s name have to appear?
In the headline, in the first sentence, and in every sentence written to be stolen — at least three self-contained factual sentences, each using the full name rather than she, he, they or the company. Pronouns die on extraction. Repeat the name where a model would otherwise lose the subject.
Does Cited invent facts if the brief is thin?
No. The drafter writes third person, inverted pyramid, from supplied facts only. No invented quotes, customers, numbers, funding, dates or places. Thin facts produce a thin release, which is safer than a confident lie on a masthead. The named person edits every word before it ships.