Claude Code for AEO: how I let it run my AI search to-do list

To use Claude Code for AEO, give it your AI search to-do list and let it do the work: I connected my AEO tracker to Claude Code with an MCP server, so one prompt has it work through every “write a page” and “get featured” task and update our website. A daily scheduled task also turns each new YouTube video into a blog post and requests Google indexing for it, with no input from me.

Last week OneAway ranked number two on Grok for “best cold email agency for B2B SaaS”. I got there with an AEO tool I built myself. AEO is answer engine optimization: getting ChatGPT, Claude, Gemini and Grok to name you when your buyers ask them a question.

The tool tells me what to do. The part that changed my week is that I no longer do most of it. Claude Code reads the task list, makes the changes to our website and ships them. This guide is the written version of the video above: how that setup works, and how you can copy it without building an app of your own.

What my AEO tracker hands Claude Code

Quick context if you’re new here. I started OneAway a little over four years ago. We’re an agency, but we now sit somewhere between an agency and a software company, because we build a lot of our own tools for ourselves and our clients. The AEO tracker is one of them. It’s a tab in our app.

Here’s what it does:

• Four engines — It sends API calls to ChatGPT, Claude, Gemini and Grok.
• 53 buyer questions — It asks the questions I think our ideal customer would ask.
• Three times per question — It asks each question three times and combines the answers, because the engines don’t cite the same sources every time.
• Two kinds of run — A daily run covers the top 10 questions. A weekly run covers all 53.
• What it records — Which sources the engines cite, whether we rank or get mentioned, and which other vendors and websites come up.

From those answers it generates four kinds of action: write a page, get featured, fix a fact and join the conversation. I explained each one, and how the tracker took us from 7.1% to 42.9% of Grok answers, in how to get recommended by AI. This post is about the next step: getting the actions done without me.

Give Claude Code your task list with an MCP server

A write-a-page action is what it sounds like. It asks you to write a page, or quite often to edit one you already have. My to-do column used to be full of them, and each one went like this: click the task, copy its copywriting prompt, open Claude Code in my website’s folder, paste the prompt and let it work.

For a non-technical person like me, that already felt amazing. It was still primitive. I was the copy-and-paste step between two pieces of software.

Some background on how our website works. The code lives on GitHub, and Vercel picks it up from there and deploys it. Everything I change in Claude Code goes to GitHub first, then Vercel puts it live. So Claude Code could already edit the site. It just couldn’t see the tasks.

So I opened our app’s folder in Claude Code and asked it to create an MCP server, because I wanted to do all of this without opening the app. It did. MCP is the Model Context Protocol, which Claude Code’s documentation describes as “an open source standard for AI-tool integrations”. In our app, you give a key your name and a label, create it, and follow the guidelines on the page to connect the server to Claude Code, Cursor or another tool.

If you build your own, Claude Code’s MCP documentation shows how to add a remote server with an API key: claude mcp add --transport http <name> <url> --header "Authorization: Bearer your-token".

Let it work through the write-a-page tasks

With the MCP server connected, the whole job is one message in Claude Code. Mine was roughly this:

Use our OneAway MCP server. Review all my AEO tasks that involve writing a page or getting featured, and take care of them.

I tried it on the day I recorded the video, because I wanted to prove it works. It found every task that needed work on our blog and handled them without my guidance. I also asked it to make an artifact, which is Claude’s way of showing what you’re working on visually. Mine showed each page it touched and what it did: added a new fact, or kept or improved what was there.

I still think write a page is the lowest-hanging fruit in AEO. Now it costs me one prompt.

Turn every YouTube video into an indexed blog post

YouTube videos are part of how we compete in AI search. A video is an asset, and it would be a shame if it only lived on YouTube.

So I set up a scheduled task in Claude. In the Claude app you open Scheduled, click “New task”, choose “Create with Claude” and describe what you want. Anthropic’s help centre says a task can run hourly, daily, weekly or on weekdays, and that it has the same access as a normal task, including connected tools, skills and installed plugins. Scheduled tasks are on Claude’s paid plans.

Mine is called “Daily YouTube blog post plus Google indexing”. Every day it checks:

1. Did I post a new YouTube video?
2. If so, is there already a blog post about it?
3. If not, it reads the transcript, writes the post for our website and publishes it.
4. Then it requests indexing for the new page in Google Search Console.

When I say “I” there, I mean Claude. I don’t do any of it. You’re reading the result. The video itself was edited by Claude Code too, animations and all. I wrote up that setup in how I edit my videos with Claude Code.

Why YouTube matters for AEO: the tracker’s join-the-conversation actions sometimes point at YouTube videos, and they don’t have to be big ones. One video Gemini was citing had eight views and had been posted seven days earlier. Small videos get cited.

Build your own AEO tracker in 3 steps

Our setup is a bit unusual, because I built a web app for myself, connected it to Claude and keep adding features. You don’t need your own app. You need three things.

Step 1: Get API keys for the engines you want to rank on

Sign up for API access to every AI engine you care about. That might be Grok, Gemini and Claude, or DeepSeek and Kimi as well. This is the first step, because everything else runs on those calls.

Step 2: Decide which questions to rank for

We generate questions from a profile we build for each client, so we know who they’re targeting. For us it’s easier, because cold email and outbound are our bread and butter. I’m also testing a feature that checks Google demand: what people actually search for is a good guess at what they’ll ask an AI.

Some ideas you can try:

• Use your clients’ own words — Give Claude access to your email, Slack and call transcripts with current clients, and ask it for the most common questions, struggles and pains they bring up. Turn those into questions.
• Add competitor questions — Take the competitors you’re up against, for example us versus Belkins, a big agency. You need enough content online to show the engines why a buyer should pick you over them. These questions are common, and you should rank for them.

Step 3: Ask every engine, three times, and turn it into a report

I don’t know a way around this one: it costs money. Make an API call to every engine for every question, ideally three times, because the engines won’t cite the same sources every time. Combine the answers for each question and build a report.

Tracking on its own is worth nothing. The API spend only pays off if the report ends in a list of actions you, or Claude Code, can work through.

If you want a shortcut: give Claude Code the transcript of this video and tell it you want to build something similar. I bet it does a pretty good job of copying what I did.

Why I still want a dashboard

There’s a view going around that software is dead and everything will happen in a chat window. I disagree. A lot of the work does happen in chat now, but I still want graphs. In my dashboard I can see at a glance that ChatGPT names us least, Grok names us most, and where we stand against competitors. It tells me whether I’m on the right path.

You don’t need much to build one. These are the tools I use:

• Claude Code — Writes the code for the tracker and the dashboard.
• GitHub — Holds the code.
• Vercel — Deploys the app so you can see it in a browser.
• An MCP server — If you want to go a step further and never open the app to do the tasks, ask Claude Code to build one.

Next I’m working on getting featured on the blogs the engines cite, without touching anything myself. I’ll show that in the next video.

Frequently asked questions

Key takeaways

• An AEO tracker is only useful if it ends in a list of actions. Mine produces four: write a page, get featured, fix a fact and join the conversation.
• An MCP server on the tracker lets Claude Code read those tasks and do them, so one prompt replaces all the copying and pasting.
• Keep your website on GitHub with Vercel deploying it, and Claude Code can ship the changes itself.
• A daily Claude scheduled task can turn every YouTube video into a blog post and request indexing for it.
• To copy it: get API keys for the engines, pick the questions your buyers ask, ask each one three times and turn the answers into a report.

If you’d rather buy a tracker than build one, see our ranking of answer engine optimization tools. For more on running work through Claude Code, read how Claude Code agent teams coordinate work.