Gregory Graf

What the work taught me

Most of my career comes down to a single hard problem: make content that reaches a lot of people and is accurate at the same time. In SEO, reputation management, and political consulting, you would be surprised how rarely those two things show up together. Plenty of tools can produce volume. Very few produce volume you would put your own name on.

I learned early that everything I published was mine to answer for. When your work is highly visible and attached to a real person, accuracy stops being a nice-to-have. It becomes the job. I got good at writing that was sourced, checked, and defensible, because that is what holds up when people are actually reading and paying attention.

Three decades the hard way

I am not a developer. I spent more than 25 years doing SEO and content marketing for multi-billion-dollar brands. Eventually I was recruited to lead SEO and online reputation management for a global consumer-products company with international operations. The job was getting ahead of Google by blending PR, earned media, microsites, and content at scale, usually with a team of one or two people.

I spent time working in Asia, cleaning up search results in languages I could not read, in countries with strict laws that ruled out content removals and anything gray-hat. The only tools available were accuracy, patience, and better content. Solving those problems over a decade is where I refined the process that everything since has been built on.

Going out on my own

In 2023 I left to start my own reputation and PR consulting firm, Snake River Strategies, which grew to include political work. I landed high-profile clients whose names showed up in the New York Times and the Wall Street Journal. Usually that was the bad kind of coverage. That is why they were calling.

Where the AI tools failed me

As the client work grew, I leaned hard into AI to scale my content drafting. I knew how to optimize for Google, and later for AI search, because reputation work means living on whatever search is turning into next. What I did not have was a tool that could write something I would put my name on.

I built spreadsheets full of prompts. I pushed ChatGPT and Claude to the edge of my patience. I evaluated the expensive tools, Jasper, Surfer, the whole shelf. They all optimized for the same things: a generic brand voice, filled-in templates, and enough keyword variations to choke a page no real person would finish reading. None of them fit a process where the work had to sound like a specific person, get the facts right, and show me the sources.

So I built it

I started experimenting, then realized I had stopped tinkering and was doing real R&D. After about three months of long days, I had something I was proud to use for my own clients and to publish my next Idaho exposé with. Ghosts.app is the result: a multi-agent AI ghostwriter where a planner researches and structures the piece, a writer trained on a real voice drafts it, an independent agent fact-checks every claim against real sources, and an editor strips the patterns that make AI writing smell like AI writing. What comes back is a draft with inline citations you can check, in a voice that actually sounds like the person it is for.

What I believe about the work

Every feature in Ghosts is tied to something I learned in the field. Citations on everything, because a claim you cannot back up is a liability. Fact-checking on every draft, because "the model usually gets it right" is not good enough when your name is on the page. Voice training, because content published under a real person has to sound like them. Per-client isolation, because the fastest way to lose a client is their voice bleeding into someone else's work. And transparency on cost and sources, because I built this assuming the person using it is accountable, by name, for every word that goes out.