Someone types a question into ChatGPT or Google, gets a full answer right there on the screen, and never clicks through to the site that answer came from.
That’s why a lot of your clients are watching their traffic drift down even though their rankings are holding steady. And it’s why the renewal call gets awkward, because you end up explaining that the rankings held while the visits fell, and you don’t have a great answer for it.
What is GEO, exactly?
Generative engine optimisation, or GEO, is the work of getting a client mentioned inside those AI answers rather than ranked underneath them.
It’s a different job from SEO, but it is not a replacement for it. Google’s own guidance is pretty clear on this: AI Overviews and AI Mode pull from the same index, the same crawl and the same quality systems as the blue links, so a page that can’t rank can’t ground an answer either. GEO sits on top of the SEO work, it doesn’t stand in for it.
What’s genuinely new is that the pool of pages an AI can cite goes a long way past the top ten. Strong, specific content sitting on page three can get quoted ahead of thin content in position one. Depending on how good your client’s content is, that’s either the scariest sentence in this article or the best one.
The exercise is making sure the model knows who your client is, what they do, and that they’re worth quoting.
Hardly anyone is doing it properly yet, which is exactly why it’s easy to sell right now.
Here’s how I’d package it, how I’d deliver it, and what I’d put in the monthly report.
Four questions that tell you if there’s a deal
Ask these on a discovery call, or bring them up with a client you already have, and you’ll know within twenty minutes whether there’s anything here.
1. When you ask ChatGPT or Perplexity for the best providers in your client’s industry, does their business show up?
This one does most of the work, because a client on the call usually goes and checks while you’re still talking.
2. What does their business own that a competitor couldn’t copy?
If this is a new client, gather as much as you can about their unique value proposition early, so you have something to build content from.
If you’re wanting to sell to a client you already have, dig that information out, and the next time you have a call with your client, ask them if they have any new products or services coming and just get a feel for whether everything you have is still true.
3. Which publications and forums do your client’s buyers read before they sign?
Next time you sit down with a client, ask them what websites their buyers frequent and see if they know.
An AI model decides whether a web agency or company is any good by checking if other companies talk about it. So make a list so you can get your client mentioned on websites they don’t own.
But there are millions of websites. You need to know which handful actually matter for this particular client.
4. Do your clients’ service pages answer buyer questions directly, or do they open with positioning statements?
Either way, you’ve just scoped a rewrite.
How do you package GEO as a service?
Tier 1: the readiness audit
This is a one-off engagement. You sit down and work out what the different models actually say about your client.
- Does their name come up at all?
- Where do they get mentioned around the web without a link?
- Can the AI crawlers even reach their site?
This report is your first deliverable. Price this low enough that a nervous client says yes on the call, because the real value of the audit turns up later. I’ll come back to that at the end.
Tier 2: the add-on
This one runs monthly, bolted onto an SEO retainer or care plan they’re already paying for. You rewrite the pages AI quotes the most (services, about us, pricing), fix the technical side, and check the answers every month.
It’s the easiest upsell of the three, because it doesn’t need a new budget approved. You’re adding a line to a plan they already signed off on.
Tier 3: the generative search retainer
This one runs separately on its own budget and includes everything in tier two, plus the two jobs that need the client themselves.
Job one: publish content only your client would know. Their buyers put questions to ChatGPT before they buy. Go after the ones no competitor has answered publicly.
- What does this cost? Every competitor says “contact us for a quote”, so the model has no numbers to work with. Put your client’s real ones on the page and suddenly it has exactly one source to cite.
- What do I actually get for that? One project, start to finish, with the timeline and the invoice.
- The question they get on every sales call. They’ve answered it out loud for years and never once written it down.
Job two: get their name onto sites they don’t own. Reddit, industry forums, comparison roundups, podcasts. This one never finishes, which is exactly why it’s monthly.
How do you actually deliver GEO?
1. Let the right crawlers in
This is one of the most common reasons people aren’t getting cited: the crawlers that feed the answers simply can’t get to the site.
And here’s where most people get it wrong, me included until I looked properly. AI crawlers are not one category. There are three of them, they do completely different jobs, and the block that protects your client’s content from one quietly deletes them from the other two.
Search and index crawlers are the ones that matter for citations. OAI-SearchBot feeds ChatGPT’s search, Claude-SearchBot supports Claude’s web answers, PerplexityBot feeds Perplexity. These build the indexes the answers get assembled from. Block them and your client’s pages can’t be retrieved, and if they can’t be retrieved they can’t be cited. This is the group people block by accident and then wonder where they went.
User-triggered fetchers fire when an actual human inside one of those products opens or asks about a page. ChatGPT-User, Claude-User, Perplexity-User. This is the closest thing you get to a real visit. Let them in.
Training crawlers collect text to train models and send nothing back. No traffic, no citation, no click. GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, CCBot and a few others. This one is a genuine business decision rather than a technical default, and it depends on whether your client’s content is their product. There’s a fair argument either way.
So the sensible default for most businesses: allow the search and user-triggered agents, then make a separate, deliberate call on training crawlers and write down why. Somebody will ask you about it in six months.
One more thing, and it catches people out. Whatever robots.txt says, the CDN gets the final word. I’ve seen sites where the robots file is wide open and Cloudflare is quietly challenging every one of those bots at the edge. Check the server logs and confirm they’re actually getting 200s back, because the robots file is a request, not a guarantee.
Then check that the actual words on the page are sitting in the HTML, because plenty of crawlers won’t hang around while JavaScript runs. Speed matters for the same reason. If the site is slow, the crawler gives up partway through and you never find out it happened.
2. Front-load the answer
These tools lift a passage or two out of a page and quote it on its own, so your job is making each section stand up without the rest of the page holding it upright.
In practice that means the question becomes the heading, and a complete answer of roughly 40 to 60 words sits directly underneath it, before any of the hedging. A model reading the page then has something usable straight away, without having to stitch four paragraphs together first.
From my experience fixing this for people, I can tell you the answer is almost always in paragraph four, buried under a story about the founder’s journey. The model doesn’t go looking for it.
I’d also save your client money on the specialist stuff being sold around this. Google has publicly knocked down most of it: llms.txt files, reformatting your content for machines, AI-specific markup, rewriting pages “for LLMs”. Writing a page that answers its own headings clearly is just good writing, and it’s doing the same job those services claim to do.
The other half is giving it something it can’t get anywhere else. Usually that’s the client’s own data, a process they’ve named and own, or numbers from their actual book of business. If a model could have written your paragraph itself from a general scrape of the web, it has no reason to cite you. It already knows what you’re telling it.
One small formatting note that matters more than it should: comparisons and step-by-step processes need to sit in real lists. Prose buries them, and these tools struggle to pull them out cleanly.
3. Make the site machine-readable
Structured data is the code in the background that tells a machine what it’s looking at. Most sites have some of it, and most of them have it in a state that teaches a model almost nothing.
The problem is that it usually arrives via a plugin, which will happily output a block describing the company and a separate block describing the article, with nothing connecting the two.
What you want is one connected set: the company, the page, the author, and a handful of standalone question-and-answer pairs, all pointing at each other. The company entry links out to every profile and listing that already exists for that brand elsewhere, which is what tells a model that the company on this page and the company in that directory are the same company.
The connections are the point. It’s worth building this by hand once per client and then maintaining it, which also happens to be exactly the kind of boring recurring work that justifies a retainer.
4. Get the rest of the web to agree
These tools check facts by looking at what everyone else seems to say, which means a lot of GEO happens on sites you don’t control.
You want the client showing up as a real presence on Reddit, Quora and whichever industry forums their buyers actually read. That’s where the model goes looking for proof that a company is any good.
Alongside that you’re chasing mentions in the publications that get indexed heavily, so comparison posts, “best X in [city]” roundups, industry newsletters and podcasts all count. You don’t need a link out of any of them. When a model picks a name, it’s going off where else that name turns up.
Then go back and clean up the mentions the client already has, so their name, their quotes and their product names match wherever they appear. Inconsistency here does real damage, because the model can’t tell it’s reading about the same business.
What goes in the monthly report
Rankings won’t carry this one. If a client is paying you every month, they need to see something that maps to what they’re actually buying, and a screenshot of position three isn’t it.
Four things, every month.
The answers themselves. Run the same set of buyer questions through ChatGPT, Perplexity and Google’s AI answers, and show whether your client’s name came back this time. Same questions every month, so there’s a line to watch instead of a snapshot.
Who got named instead. The competitors turning up in the answers your client didn’t make. This is the slide that renews the retainer.
New mentions off-site. Where their name showed up this month on sites they don’t own, link or no link.
What you shipped. Pages rewritten, schema built, technical fixes made, and what’s queued for next month.
Screenshot the answers. A client who sees their own name sitting inside a ChatGPT response doesn’t need the rest of the report explained to them.
So why price that audit low?
I said I’d come back to it.
Because the audit isn’t really the product. It’s three things at once, and none of them are the PDF.
It’s your baseline. Month one of a retainer is worthless without a before. The audit is the before, and you already got paid to produce it.
It’s your scope. Everything broken that you found is a line item the client has already seen with their own eyes. You’re not pitching work anymore, you’re quoting a fix for problems they watched you uncover.
And it’s your close. A client who has sat there while ChatGPT named three of their competitors and not them is pre-sold. You don’t have to explain why this matters. The answer on the screen did it for you.
That’s why you want it cheap enough that nobody has to think about it. The audit buys you the retainer.
We’re making this easier for you
Selling the audit is the easy bit, you can do that anytime.
We’ve been building Atarim to take a lot of the rest off your plate.
It gives you a team of AI experts working across every site on your list instead of one at a time. Index, the SEO and performance one, goes through the whole client list for structure, metadata and crawlability and tells you what’s broken.
Lexi, the content one, drafts the page rewrites in the client’s own voice, so the money pages burying the answer in paragraph four come back to you already fixed.
Nothing moves until you approve it. You set what waits for you and what just runs. Click here to add your URL and give it a spin.
So go make some money and build a stronger lasting relationship with your clients, and we’ll be here if you need any support.
Which client are you checking first?
