People don't scroll ten blue links anymore. They ask ChatGPT "what's the best X for Y" and take the answer. Same with Perplexity, Gemini, Copilot, Google's AI Overviews. One answer, a couple of brands named, done.
If your brand is one of the names, great. If it isn't, there's no page two to claw your way up to. You're just not in the room.
I've been building FixAEO around that problem for a while — tracking whether AI engines cite you, and where they cite your competitors instead. But tracking is the easy half. The hard half is doing something about it every week: rewrite this page, refresh that stale stat, figure out which of your pages the models actually pull from, then do it all again next Monday. That part is a job.
So I built the thing that does the job. It's two pieces: an AI Marketer (recipes that do the work for you) and an agent builder (build your own when the recipes aren't enough). This is the honest walkthrough of both — including the parts that aren't finished.
Quick definition for the devs who haven't touched this yet: AEO is answer-engine optimization. It's SEO's weird younger sibling — instead of ranking a page, you're trying to be the source an LLM quotes when someone asks it a question. Different game, same stakes.
The AI Marketer: recipes that run on your real data
The important word is real. This isn't a chat box that hands you generic "write more content" advice. It runs on your actual citation data — the prompts we run across six AI engines (ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, Google AI Mode) to see where you show up and where you don't.
The AI Marketer mid-run on the Weekly Brand Health recipe — note the confidence badge and the "review generated work before use" footer.
It ships with 17 ready-made agent templates. The five I reach for most are the core content and report recipes:
- Content Optimization Suggestions — point it at a page and a question you want to win, and it gives you source-linked fixes.
- AEO Content Refresh — reviews the recommendations, then generates a before/after draft.
- Content Brief + Final Draft Generator — researches a topic, builds a brief, then writes a sourced draft.
- Optimize Lowest Performing Cited Page — finds your least-cited owned page in a scope you set and tells you how to fix it.
- Weekly Brand Health Report — compares two full weeks of your mentions and citations so you can see what moved.
The template gallery — ready-made agents across content optimization, creation, research, reporting, and outreach.
Two of those (AEO Content Refresh, Content Brief + Final Draft) deliberately stop and wait for you before they generate the final draft. You review the plan, then it writes. I didn't want a recipe quietly shipping a draft you never looked at.
Every recipe shows its plan and deliverable up front — and states plainly that nothing is published automatically.
The part I'm actually proud of is the grounding. The recipe engine won't let the model invent search volumes, competitor numbers, or outside facts. Every quote and citation it uses is machine-checked against the sources it actually pulled — anything that doesn't match a captured source is rejected. For the recommendation recipes, unbacked items get dropped; for the draft recipes, it re-runs once and fails the step rather than ship something ungrounded. LLMs are confident liars by default; this makes them cite their work or shut up.
And you don't have to babysit the Weekly Brand Health report — put it on a weekly schedule and it just runs. The other recipes run on demand, whenever you want a fresh pass. (The full eight-frequency scheduler lives in the agent builder, below.)
Cost model: work credits. One credit is one cent of measured model cost. The $29 plan includes 100 credits a month. You spend what the work actually costs — a reply might burn zero credits or a few, depending on what it did. No made-up "premium action" tax.
The agent builder: for when the recipes aren't your shape
The recipes are the paved road. The builder is for going off it.
It's a node graph. You wire up the steps yourself: pull your brand's data, run a skill, hand off to an integration, gate on a human review. Up to 10 skills per agent — and that's enforced on the server, not a polite suggestion in the docs.
The integrations that actually exist right now (I'm only listing what's shipped):
- Slack — have an agent post into a channel
- Google Docs — drop a draft straight into a doc
- WordPress — publish a finished draft straight to your blog as a draft post
- Webhooks and a generic HTTP API node — POST to your own endpoint and do whatever you want with it
Eight schedule frequencies — hourly, daily, weekdays, weekends, weekly, monthly, annually, or just once — so an agent can wake up every weekday morning, do its thing, and drop the result where you'll see it.
The safety model, stated precisely
This is where a lot of "autonomous AI" pitches get slippery, so let me be exact.
Where you place a review checkpoint in an agent, the run stops there. It cannot be skipped — not even in a test run. And for anything that sends externally or is destructive, you're not allowed to set "always allow" — those get a human every time, and each approval is single-use, so one click doesn't hand over the keys forever. (Low-risk internal steps you can choose to remember; the irreversible stuff you can't.)
A run paused at a human-approval step. Approving continues with those exact values; rejecting stops the run — and external actions get one-time approval only.
Here's the honest caveat: the builder does not force you to add a review step. If you wire up an agent that sends something with no checkpoint, it'll warn you — but it'll still run. So the guardrail is real and it's strict where you use it, but it's opt-in, not mandatory. I'd rather tell you that than pretend the thing is idiot-proof. It isn't. It's careful, if you build it careful.
The honest part (a.k.a. what it doesn't do)
I'm building this in public, so:
- It's early. I'm still shipping reliability fixes to the builder as I write this. If you throw a weird graph at it, you might find an edge I haven't. Please do.
- $29 scans refresh every 72 hours, not daily. If you need daily freshness, that's the $79 tier. I'm not going to bury that.
- $29 is one seat, 15 tracked prompts, up to 2 brands. It's built for a solo founder or a small team watching a couple of brands — not a 40-person agency. Right tool, right size.
- There's no giant connector marketplace. What's shipped: Slack, Google Docs, WordPress, and a webhook/HTTP node. That's it for now, and I'd rather ship a handful that work than name twenty that don't.
Why $29 is the number that matters
Not because it's cheap for cheap's sake. Because the agent builder and the AI Marketer aren't locked behind some "call sales" enterprise tier. $29 (or $25/mo if you pay yearly) is just the lowest plan where they have real citation data to chew on — six engines, actual mentions, actual competitor leaderboards. There's also a genuinely free tier — a one-engine (Gemini) scan — if you just want to check whether AI knows you exist before you spend anything.
The plans. Everything I described lives on the $29 Lite tier.
Most tools doing agentic AEO work sit a good deal north of $29 a month. I won't run a teardown here — the pricing pages are all public, go compare them yourself. That's kind of the point.
Go break it
I'm not going to hard-sell you. Do one thing: run a free scan on your own domain and see if AI even names you. If it does, nice. If it doesn't, that's the gap that's quietly costing you.
And if you build an agent that does something dumb, tell me. Right now that feedback is worth more to me than the $29.
Top comments (0)