The objection we hear after every audit
Can you do GEO yourself with AI after an audit?
The short answer
You can, and almost nobody does. Knowing what to write is the easy part. Writing it every week, with a method, is the hard part. Three weeks after an audit, most teams that planned to do it themselves with AI have published nothing.
Published September 29, 2026 · based on the AI visibility audits we delivered in September 2026 · written by Thibault Genon-Catalot, founder
In September, I delivered AI visibility audits to several companies. On almost every call, someone said the same sentence: “Thank you, I’ll give this to Claude and it will write the pages.”
It sounds right. The audit tells you what to write, and AI writes fast. So why pay anyone to do it?
As someone who helps businesses create GEO content, I am not thrilled to hear it. I am also a little embarrassed for them, because so much is wrong with this way of thinking. From what my clients tell me, three weeks later most of them have published nothing.
Knowing what to write is the easy part. Writing it every week, with a method, is the hard part. That brings me to the two least sexy words in marketing: methodology and consistency. Here are seven reasons to hand the implementation of your GEO strategy to your agency.
Seven reasons
Why doing it yourself with AI rarely happens.
Execution never happens
Three weeks from now, you know deep down that nothing will be done. If you don't know it, I do.
Consistency is what makes this work. You publish and follow up every week for 12 weeks, while your marketing lead has 15 other priorities. Most teams write three pages, stop, and decide two months later that it didn't work.
The review bottleneck
CMOs and CEOs are busy people. They want to review everything before it is published, on LinkedIn or on their website. When they are also the ones creating the content, they become their own bottleneck.
An agency takes care of the writing and the proofreading. You spend about 2 hours a week making the decisions only you can make, and approving what is published in your name.
Volume
GEO pays only in volume. The engines start moving when they find you on many sites and many pages. Doing GEO seriously means industrializing content production.
Three articles a month for three months is peanuts. Nine articles a week is a good start. Volume matters, and as volume goes up, quality goes up too. Your competitors keep publishing as well, so the longer the gap stays open, the more pages it takes to close it.
A GEO article is not an SEO article
Most companies I work with barely know how to write an SEO article. They assume Claude will naturally write a GEO article because it is an AI itself. It doesn't work that way.
When did you last write a GEO article? It is a different job, and it needs a proper method. Agencies know the formats that the engines pick, and they run health checks before a page goes live. The field is also two years old and changes every month. Keeping up takes constant attention to how the AI engines evolve.
The AI tooling itself
Most people I talk to think they can hand everything to Claude and it will be fine. Most of them have never built a robot, an agent or even a website with it. Three small experiments do not make a tool that works.
I have built many apps, automations and robots. Nothing that works is easy. It takes a method, a long time of testing, patience and a lot of good ideas, because nothing is a miracle with AI. How long would it take you to build your own robot for something that is maybe 5% of your work? You will never do it, and if you try, it won't be a good tool.
AI visibility agencies have their own tools because this is their job. They use them every day. Building your own is neither a reliable solution nor a good use of your time and money.
Beyond the pages
GEO is not only about creating pages. A large part of the effort goes into being visible on third-party sites: lists, articles, forums and social media. Getting added to a list takes a request, a follow-up and another follow-up. As a CMO, you don't have time for that.
The report is the tip of the data
Most businesses don't realize that the audit report is an HTML or PDF dressing of the raw data. The client reads a fraction of it. The rest is either noise, or data they don't know exists and will never see. Whoever implements with all the raw data can use far more of it.
Our own numbers
What an audit report shows, against what the audit collected.
8 AI visibility audits we ran in September 2026, about 300 AI answers each (20 buyer questions, 5 engines, 3 runs). “In the report” means printed in the report’s tables. Counted on September 29, 2026.
| What the audit collected | Found | In the report | Share |
|---|---|---|---|
| Brands named by the engines | 3,907 | 335 | 9% |
| Websites used as sources | 4,084 | 765 | 19% |
| Individual pages cited | 9,104 | 1,394 | 15% |
A typical audit finds around 500 brands named by the engines, and the report’s tables show about 40 of them. It finds around 500 websites used as sources, and the tables show about 1 in 5. It finds more than 1,000 individual pages cited, and the tables show about 1 in 7. The audits also collected 20,800 citation links and the full text of every answer. The same goes for every other kind of data in the audit.
Conclusion
Only the companies that are really willing to invest in their GEO strategy will actually do something about it. You have no idea how many times I have heard “it’s fine, I’ll do it with AI”, and three weeks later nothing has happened.
The people who plan to build their own AI robot are often the ones who have never written a GEO article themselves. They have no method, and they have never been consistent with content. Most of them know they are busy. Few are honest enough to admit the limits of what they can do alone.
Questions
Doing GEO yourself. Straight answers.
Can I really do GEO myself with ChatGPT or Claude?
Yes, it is possible. The audit tells you what to write and the AI writes fast. What breaks is the rest: publishing every week for 12 weeks, reviewing every page yourself, writing for the engines rather than for Google, getting added to outside lists, and working from the full data instead of the report. In our experience, three weeks after an audit most teams that planned to do it themselves have published nothing.
How many pages does GEO need?
More than most teams expect. The engines move when they find you on many sites and many pages. Three articles a month for three months is too little to show up; nine a week is a good start.
What does an audit report leave out?
Most of the data. Across 8 audits we ran in September 2026 (about 300 AI answers each), the reports' tables showed 9% of the brands the engines named, 19% of the websites they used as sources and 15% of the pages they cited. The rest stays in the raw data.
We run the implementation for you in the AI Visibility Program, from the full data of your audit. The method is public at /methodology, and our own study of 615 AI answers is at /do-ai-answers-change-every-time.