AI Press Release Polisher
Give the announcement a real news lead while preserving names, dates, figures and quotations. Try the same tool available in your dashboard, then carry your input, options and result into your account.
Give the announcement a real news lead while preserving names, dates, figures and quotations.
Add at least 25 words to draft release and facts.
The finished release is formatted as a newsroom document with protected facts and quotes.
Five things a journalist checks, in the order they check them.
Most of the decision happens in the first sentence, which is why most of the editing should happen there too.
- Whether there is news in the first sentenceThe whole decision usually happens here. A journalist opening a release is asking one question: is there a story. If the first sentence is about your company being pleased, the answer arrives as no before the news does. Put what happened in the opening sentence — what changed, for whom, when — and let the context follow.
- The corporate phrases that carry no information“Is pleased to announce”, “leading provider of”, “innovative solution”, “strategic partnership”, “commitment to excellence”. Every one of these is a claim nobody can check attached to a sentence that would be stronger without it. Editors delete them on sight, so a release full of them arrives pre-shortened and reads as though nobody edited it.
- Whether the quote says anythingMost press release quotes are the same quote: an executive expressing excitement about a milestone. It survives because nobody wants to write a real one, and it is dead weight. A usable quote says something only that person could say — why this was hard, what they expected and got wrong, what it means for a customer. If your quote could be attributed to anyone at any company, it will not be used.
- Numbers, dates and names that can be checkedThe thing that separates a release a journalist can work with from one they cannot. How many, since when, in which markets, compared to what. Vague scale claims are worse than no claim, because they signal there is no figure worth quoting. And if a model drafted any of these, verify every one before it goes out.
- Where the boilerplate sitsThe paragraph about your company belongs at the bottom, after the news, where somebody writing the story can find it. Opening with it is the most common structural mistake in the format and it costs you the reader before the announcement.
Why this edits your draft rather than writing one.
Nearly every tool ranking for this term will write a complete release from a short brief. That is a genuinely different product from this one, and the difference matters more here than in most kinds of writing.
The facts in a release exist only inside your company. How many customers, since when, in which markets, what the executive actually thinks. A model asked for a complete announcement will produce all of those, and it will produce them plausibly, because plausible is what it is good at. A release goes to a distribution list and onto the record, so an invented figure is not a draft problem — it is a correction you issue publicly.
Search-based tools introduce a subtler version. A drafted release can arrive containing claims about your market or your competitors that came from its sources rather than from you, phrased as though your company is asserting them. Anything comparative needs a source you can produce on request.
What a model is genuinely good at is the shape. Recognising that the news is buried in paragraph three, that the quote says nothing, that four phrases are doing no work. Those are editorial judgements about a document that already exists, and they are most of what a weak release needs.
So the split here is deliberate. You supply what happened. This fixes where it sits.
Releases drafted with ChatGPT, Claude, Gemini, DeepSeek, Grok, LLaMA and Perplexity.
ChatGPT press release polisher
It produces the format flawlessly, including the parts of the format that do not work.
ChatGPT has read an enormous number of press releases and reproduces the conventions exactly: the pleased-to-announce opening, the executive quote about excitement, the boilerplate, the contact block. It is a competent example of the genre, and the genre as commonly practised is what editors have learned to skim.
The useful edit is inverting its structure. The news is usually in ChatGPT's third paragraph, because it set up first. Move that sentence to the top, delete the paragraph that preceded it, and most of the release improves without another word being touched.
Claude press release polisher
The most readable prose of any assistant, and the most cautious.
Claude hedges, which produces a release where the announcement is qualified into vagueness. For a regulated announcement that caution is appropriate and worth keeping. For a product launch it means no journalist can find the claim, and a story needs a claim.
It also arrives with Markdown structure — bolded lead-ins and headed sections — which does not survive a paste into a distribution system or an email and shows up as literal characters in what an editor receives.
Gemini press release polisher
Gemini fragments a release into a briefing document.
It returns headings and bullets, so an announcement arrives as a summary of itself. Releases are prose for a reason: a journalist is looking for sentences they can quote or adapt, and a bullet is neither. A bulleted release also signals that nobody in the process wrote it as a story.
Its output travels through Workspace more often than the others', so formatting artefacts persist further into the distribution chain than you would expect.
DeepSeek press release polisher
Technically precise, and technical is not the register.
Its output reads as documentation, which is right for a technical annexe and wrong for a release meant to be read by a general-assignment reporter. Specification detail belongs below the news, not in place of it.
Reasoning-model output also frequently still carries the reasoning, which in a public announcement is a genuine risk. A release goes to a distribution list and cannot be recalled, so check the top and tail of anything drafted this way before it leaves.
Grok press release polisher
The register is too informal for the format, which is occasionally the point.
Grok writes conversationally, and a press release is a formal document read by people who scan hundreds. Most of the work is raising the register while keeping whatever was actually direct about the draft, because directness is the one thing most releases lack.
Where the informality helps is the quote. A Grok-drafted quote sounds more like a person talking than the standard excitement boilerplate, and it is worth keeping the shape of it while making the content true.
LLaMA press release polisher
The case where facts need checking most carefully.
LLaMA is weights rather than a product, so output varies with the deployment. Smaller ones drift on specifics — a figure that moves, a date that shifts, a name spelled two ways in one document. In a press release those are not stylistic issues; they are corrections you will be issuing publicly.
Verify every number, date and name against your own source before distribution, whichever model drafted it, and doubly so here.
Perplexity press release polisher
Useful for the market context, risky for the announcement.
Because it searches, a drafted release can quietly include claims about your market, your competitors or your sector that came from its sources rather than from you. Those read as though your company is asserting them, which is a legal and reputational exposure rather than an editorial one.
Anything comparative — market position, growth relative to a sector, being first or largest — needs a source you can produce on request. If you cannot say where a claim came from, it does not go in the release.
Press release polisher questions.
What does a press release polisher do?
It edits a release you already have rather than writing one from scratch: the news moved into the first sentence, corporate filler cut, the structure inverted so the announcement precedes the context, and the boilerplate moved to the bottom. Every fact, figure, date, name and quote comes through exactly as written, because those are yours.
Why not just generate the whole release?
Because the facts in a release come from your company and nowhere else, and a model asked to produce a complete announcement will fill the gaps with plausible material. In a document that goes to journalists and onto the record, an invented figure or an unsupported market claim is a considerably worse problem than a clumsy sentence.
What is wrong with “is pleased to announce”?
It puts your company's emotional state where the news should be, and it delays the actual information by a clause. Editors delete it automatically. More importantly, a journalist deciding whether to read on gets no reason to in the first six words, which is where that decision is made.
Do journalists know when a release was AI-written?
They increasingly assume it, which is the more useful way to think about it. Newsrooms are receiving far more releases than before and most of them read as generated, so the assumption costs you nothing to overcome and quite a lot to confirm. A specific first sentence and a quote that only one person could have said are what distinguish a release, not whether a tool touched it.
How do I write a quote worth using?
Say something only that person could say. Why it was hard, what surprised them, what changes for a customer on Monday. The default excitement quote is unusable because it is interchangeable, and an editor with limited space cuts the interchangeable thing first.
Can it polish releases drafted with ChatGPT, Claude or Gemini?
Yes, and the sections above cover what each produces. ChatGPT reproduces the format including the parts that do not work, Claude hedges the announcement into vagueness, and Gemini returns bullets where a journalist needs sentences they can quote.
Will it check my facts?
No. It preserves what you wrote and cannot verify it. Every number, date, name and comparative claim needs checking against your own source before distribution, and this matters more than usual with anything a model drafted, since search-based tools can introduce market claims that read as your company asserting them.
Where should the boilerplate go?
At the bottom, after the news, clearly separated. Somebody writing the story will look for it there. Opening with your company description is the most common structural mistake in the format, and it spends the reader's attention before the announcement arrives.
How long should a release be?
Shorter than the one you have. The news in the first sentence, the necessary context in two or three paragraphs, one usable quote, the specifics, then boilerplate and contacts. Length signals that nobody decided what mattered, and deciding what matters is the job.
Is my draft stored?
The document is attached to your account so you can return to it, and you can delete it whenever you like. It is not used to train anything and it is never distributed anywhere.