GEO Agent
Somebody asked an assistant which tool does what you built, and it named three products that were not yours. This agent keeps asking that question, records who got cited, names the page of yours that should have been, and opens the pull request that fixes it, on the repository you connect.
You do not know what the engines say about you
Your buyers stopped typing keywords into a search box. They ask an assistant which tool does the thing, and it answers with three names and a couple of citations. You have no idea whether you are ever one of them, and the honest answer is that if your robots.txt was written in 2023 and you have never served an llms.txt, you probably are not.
What is GEO?
GEO is making a site quotable by the engines that answer questions instead of listing links. When somebody asks an assistant which tool does the thing you built, the answer names two or three products and cites a handful of pages. Being one of them is a different job from ranking: the page has to state the answer plainly, be marked up so a machine can tell what it is claiming, and be reachable by crawlers that most robots.txt files were written before anybody had heard of.
A GEO agent is that job on a cycle. It asks the questions your buyers ask, records whether you were named and which competitors were, and for every question where only a competitor appeared it names the page of yours that should have earned the citation and why it did not. Then it fixes what it can fix in code: the crawlers you are blocking, the llms.txt you never wrote, the schema that does not say what the page says.
What it leaves you
No proposals: a site fix never goes through the approval feed. This agent produces a visibility probe you can read, recommendations tied to the questions you lost, and the pull requests that fix them.
The skills it runs
A run may activate these and nothing else. Every one of them is open: read the file before you turn the agent on.
What the AI CMO market charges for AI visibility
These are pages and prices on those companies' own sites on the day shown. Kwern publishes no citation-rate or visibility benchmark of its own, because the only honest number here is the one your own probe returns.
Kwern against the alternatives
| What is compared | Kwern | AI CMO suites | Point tools | An agency | Doing it yourself |
|---|---|---|---|---|---|
| Letting the engines read you at all | Checks which answer engines your robots.txt is turning away and whether you serve an llms.txt at all, then fixes both in a pull request. | A dashboard of mentions, with the robots file left as you wrote it. | A generator hands you a file. Shipping it is still your job. | They have read the same three blog posts about GEO that you have. | You have been meaning to write an llms.txt since February. |
| Finding out whether you are the answer | Runs a probe on the questions your buyers actually ask, records who was cited, and names the page of yours that should have been. | A visibility score, without the page that would change it. | A tracker counts mentions and stops there. | A screenshot of one ChatGPT answer, in a slide. | You asked it about yourself once and did not like the answer. |
| Writing the fix | The diff: the schema block, the answer paragraph, the llms.txt, written into your own files. | Prose advice you then implement yourself. | Nothing. A score is not a change. | A recommendation deck, and your developer is still you. | A schema generator, a copy and paste, and a hope. |
| Who says yes before it goes out | One feed. Approve, edit or reject each draft; nothing is published or spent until you do. | Usually a dashboard, sometimes autopilot with no stop. | Per tool, in whatever queue that tool keeps. | A shared document, and a reply a day later. | You wrote it, so you approved it. |
| What it costs a month | From $29 a month, every agent in every plan. | $20 to $1,000 a month, and $5,000 for a done-for-you tier. | One subscription per channel, each with its own seat. | A monthly retainer, agreed before anything ships. | Nothing, and it costs you the evenings. |
| Who pays for the model | Your own key. Across the 11 models Kwern vets, input runs $0.10 to $10 per million tokens, billed to you by the provider. | Credits you buy from them; the markup is never itemised. | Theirs, folded into the subscription. | Inside the retainer, at whatever margin they keep. | Your own key, at the provider's price. |
RiseMore, the cheapest
- www.risemore.app/pricing
MarketOwl, the dearest
- marketowl.ai/pricing
This Agent runs about 2 times a month. The tokens those runs spend are billed to you by your own provider: $0.10 per million in on gpt-6-luna, $10 on gpt-6-astra. Kwern never resells a token and never marks one up.
Three plans, every agent in each
Your model key, your accounts, your approval on every post. The plan decides how many products you run.
Questions about the GEO Agent
Where to go next
Someone is asking for what you built today.
On Reddit, on X, in ChatGPT's answers. Kwern finds them every morning and drafts the reply in your voice; you decide what goes out.