About Ghosts
Every ghost has a backstory.
Most AI writing tools hand you a fluent draft that is subtly wrong: a statistic that does not exist and a structure you have read a thousand times, in a voice that belongs to no one. Ghosts was built by someone who spent 25 years writing for high-profile brands and campaigns and could not ship that.
The person behind it
Ghosts was founded in July 2026 by Gregory Graf, who for 25 years has worked in SEO and online reputation management alongside political communications, writing and directing content for high-profile brands and individuals from the Boise, Idaho area. Writing for names that big teaches you two things quickly: the facts have to be right, because someone else's name is on them, and the voice has to be theirs, not a house style wearing their name.
When AI writing tools arrived, they failed on both counts: the drafts read well but fell apart under scrutiny, with numbers showing up without sources, and every client started to sound like the same competent stranger. Better prompts moved the needle a little and then stopped moving it at all, which left one practical question: what would it take to make an AI draft you would actually put your reputation behind?
Why one model is not enough
A single language model predicts the most likely next word, and across enough text that "most likely" drifts toward the average of everything it has read, which is why AI writing sounds the same no matter whose brand it is supposed to carry. Ask that one model to plan, draft, check its own facts, and hold a specific voice at once, and it trades those four goals off against each other, so the result is competent and forgettable and occasionally confidently false.
The fix is a different shape. Writing has distinct jobs that each need their own kind of attention, so Ghosts gives each job to a specialist and links them with checks. Any serious publication works the same way: the person who writes and the person who checks the facts are separate people, for a reason.
How Ghosts writes
- 1
It starts with the details, not a blank box
Before a word is written, the assignment is assembled from the things a piece actually needs: the format and length for where it will publish, the channel's optimization rules, plus a researched understanding of the brand you can read and correct first. A planning agent researches the topic from several angles and commits to a thesis and a structure. The draft is written against a plan, not improvised into one.
- 2
A writer trained on a voice, not a topic
The drafting persona learns how a specific author or brand writes: rhythm, word choice, how they open and how they land a piece. Trained on writing samples, it improves from every edit your team makes, with hygiene filters so a single odd correction does not throw the voice off. What it learned is visible and reversible, so the voice never drifts somewhere you cannot see. Pick from a bench of built-in expert writers, including a dedicated legal specialist bench, or train a custom writer on your own published work.
- 3
Checked by an agent that did not write it
An independent fact-checker checks the draft against the sources found in research, reading their actual text rather than the model's memory. Verified numbers and dates are locked so a later polish cannot quietly change them. A line editor rewrites the machine-writing patterns out of the prose, and on the top tier a fresh reviewer reads the finished piece cold. Every draft carries a log of what was flagged and changed, plus the true cost of the piece. Nothing about the process is a black box.
Who it is for
Agencies
Hold a distinct voice for every client across every channel, without re-prompting each piece by hand. The voice compounds; you do not babysit it.
In-house and enterprise teams
One fabricated statistic in a published piece is a reputational and sometimes legal problem. Verification is built into the pipeline, not left to whoever proofreads last.
Creators and consultants
Publish under your own name in your own voice, at a volume you could not sustain by hand, without it reading like everyone else's AI.
Law firms
A dedicated specialist bench and retention you control, with stricter claim auditing that never invents a citation. Built for content that has to survive scrutiny.
How to use it
Set up a project once with the brand and its audience and voice, then pick a writer from the roster or train your own on real samples. Then, for each piece, choose where it will publish and give it a topic. Ghosts researches and confirms anything ambiguous with you, then plans and drafts against that plan while checking its facts, showing its work at every step. You read the editor's notes, apply the changes worth applying, then publish. The more you edit, the more the writer sounds like you.
Common questions
Who is behind Ghosts?
Ghosts was founded in July 2026 by Gregory Graf, an SEO, online-reputation, and political communications consultant based in the Boise, Idaho area. He has spent 25 years writing and managing content for high-profile brands and individuals, and built Ghosts to solve the accuracy and voice problems he kept hitting with single-model AI tools.
How is this different from ChatGPT or a single AI writing app?
A single model plans, drafts, fact-checks, and imitates a voice all at once, and does none of them well. Ghosts splits the work across specialized agents: a planner that researches and commits to a structure, a voice persona trained on your writing, an independent fact-checker that verifies claims against real fetched sources, and editorial passes that strip the machine-writing fingerprint. Verification is done by an agent that did not write the draft and has no stake in its claims being right.
Does Ghosts just make things up like other AI tools?
That is the failure mode it was designed around. Every factual claim is checked against the actual text of fetched sources, not the model's memory. Numbers and dates that pass the check are locked, so a later editing pass cannot quietly change a verified figure. If research turns up an ambiguous subject, the system asks you to confirm which one you meant before it drafts, rather than guessing.
Will it actually sound like me or my client?
The persona is trained on writing samples, not topic knowledge. It learns sentence rhythm, diction, and how you open and close an argument. Every edit your team makes teaches it a durable lesson, with filters so one odd edit does not corrupt the voice, and the full learning ledger is visible and reversible in the product. The tenth piece for a client is measurably closer to their voice than the first.
Who is Ghosts for?
Agencies managing many client voices, in-house and enterprise content teams where a single fabricated statistic is a real liability, solo creators and consultants who publish under their own name, and law firms, which get a dedicated specialist bench and stricter claim auditing. If your writing goes out under a name someone has to answer for, it is built for you.
Is my content used to train AI models?
No. Your content is never used to train AI models. That is our policy and a contractual commitment from our AI providers. Every workspace is isolated, and nobody at Ghosts reads your drafts in ordinary operation.
Where to find us
Ghosts publishes product notes, writing craft, and detection-benchmark results at X (@aighosts), LinkedIn, and Facebook. For press or partnerships, contact us directly.
More about the founder: Gregory Graf, the SEO and reputation strategist who built Ghosts after 25 years in political communications.
Ghosts is a product of its founder's 25 years of writing for clients by hand. Writing help you would be proud to publish.