AI SEO agency or AI visibility tool: which do you actually need?
They are not alternatives. An AI visibility tool measures where you appear in AI answers and costs $29 to $250 a month. An agency does the work that changes it, from $5,000 a month or a fixed price audit. Almost everyone should buy the measurement first, because without a baseline you cannot brief an agency or judge one.
By Rajat Kapoor
Updated October 2026
Key takeaways
A tool measures, an agency changes things. Buying a tracker and expecting visibility to improve is buying a thermometer to treat a fever.
The price gap is roughly fifty to one, which is why the measurement comes first for almost everyone: a baseline costs less than one day of agency time.
Google states that no special optimisation is needed for its AI features beyond ordinary Search eligibility, and that new machine readable files are not required.
That statement covers Google only. ChatGPT, Perplexity and Claude run their own crawlers and their own retrieval, so one vendor's position is not the whole picture.
OpenAI runs separate crawlers for model training and for ChatGPT search, and blocking the training one does not remove you from ChatGPT's search results.
If an agency measures you with a tracker it owns, ask what happens to the history when the contract ends.
If your conventional search presence is broken, fix that first. The work overlaps heavily and nothing in AI retrieval rewards a site that cannot be crawled or understood.
Someone has asked you what the company's AI visibility is, and you have found two kinds of answer for sale. One is software at a few hundred dollars a month that will show you a number. The other is an agency at several thousand a month that will promise to improve it. They are marketed with the same words, GEO and AEO and AI search, and they are presented as a choice.
They are not a choice. One measures and one acts, and the useful question is not which to buy but in what order, and whether you need the second one at all. This page sorts that out, with the numbers from both markets.
What a tool does, and exactly where it stops
An AI visibility tool runs a list of prompts across ChatGPT, Perplexity, Google AI Overviews, Gemini and Claude on a schedule, and records what comes back: whether you are mentioned, in what position, whether your site is cited as a source, and how your share of those answers compares with named competitors.
That is genuinely worth having, and it is the only way to know where you stand. It is also the whole of what the software does. It will tell you that you are absent from the answer to the single most commercially valuable question in your category. It will not write the comparison page, fix the markup, or persuade the review site that ranks above you to mention your name.
One caution before you treat any of these numbers as fact: two tools will disagree about you for reasons that have nothing to do with your marketing. Our guide to what an AI visibility score actually measures explains why, and is worth reading before the figure goes in a board pack.
What the engines themselves say, which is not what the market says
This is the part that changes the decision, and almost nobody selling into this market will lead with it.
Google's own documentation on AI features in Search states that a page needs to be indexed and eligible for a snippet, and then says there are "no additional requirements" to appear in AI Overviews or AI Mode. It goes further and says you do not need to create new machine readable files or AI text files for the purpose. If you have been quoted for an llms.txt implementation as a Google visibility measure, that is the vendor's position on it.
Two things are true about that statement at once. It is authoritative, because it comes from the company that runs the system. And it covers Google and nothing else. ChatGPT, Perplexity and Claude are separate retrieval systems with their own crawlers and their own source selection, and none of them has said the same thing.
OpenAI's crawler documentation is the clearest worked example of why that distinction matters. It runs separate bots for separate jobs: GPTBot gathers content for model training, while OAI-SearchBot is the one behind ChatGPT's search feature. Plenty of sites blocked GPTBot in 2023 and 2024 to keep their content out of training, which was a reasonable decision at the time and has nothing to do with whether ChatGPT can surface them today. If someone tells you that you are invisible in ChatGPT because you blocked the training crawler, check which line is actually in your robots.txt.
Neither of those facts costs anything to establish. Both are the kind of thing a good agency settles in the first hour and a tool will never raise with you at all.
What an agency does that software cannot
Three things, in rough order of how often they turn out to be the real problem.
The first is content that answers the question directly. AI systems extract; they do not reward a page that buries the answer under six paragraphs of context. Restructuring a content library for extraction is editorial work and there is no tool for it.
The second is third party sources, and this is the one buyers underestimate most. An assistant asked to recommend a vendor leans heavily on things that are not the vendor: reviews, comparisons, discussion threads, press. You cannot publish your way into those. That is why digital PR and earned media practices are now appearing inside AI search offers, and why a purely on site programme tends to stall after the first few months.
The third is entity clarity: whether the systems correctly understand what your company is, what it sells and who it competes with. That is slow work and it is where the specialists earn their fee.
The price gap is the decision for most people
The published numbers are not close. Across the tools on our AEO tools ranking, entry pricing runs from $29 a month at the cheapest to $250 at the most expensive, with most sitting between $60 and $165. Across the firms on our AI SEO agencies ranking, published agency pricing starts at $5,000 a month for search and answer engine work together, runs to $8,500 and $10,000 a month for a full programme, and reaches $22,000 and above at the top of a published tier card.
That is a factor of roughly twenty at the narrowest and two hundred at the widest. A year of the most expensive tracker costs less than one month with most of these agencies.
There is a middle option that is easy to miss. Two firms sell a fixed price diagnostic rather than a retainer: a $15,000 AI Search Sprint and a defined AI Search Audit covering entity mapping across the five main engines, a crawler level technical check, and a visibility baseline. For a company that has never been measured, that is often the right first purchase, because it answers the question a tool cannot: not where you stand, but why.
Route by where you actually are
You have no baseline. Buy the tool. Nothing else is sensible until you can see the current position, and at these prices the decision does not need a business case. Run it for at least a month before drawing conclusions, because the numbers move week to week.
The baseline shows you are absent, and you have a content team. Tool plus in house work. Most of the first round of fixes is editorial and structural, your team can do it, and you now have the instrument to tell whether it worked. Revisit in a quarter.
The baseline shows you are absent, and you have no capacity. This is the clearest case for an agency, and it is worth paying for a diagnostic first so the retainer starts with a scope rather than a discovery phase.
You are cited, but competitors are cited more. Agency, and specifically one with an earned media or digital PR practice. This is the hardest problem in the category because the lever is outside your own site, and it is the one an in house team is least equipped to pull.
Your site is the problem. If crawlers cannot render your pages, or a replatform is coming, that is a technical engagement and a short one. Get it done before the content work, not after.
Your conventional search presence is already broken. Fix that first. The underlying work overlaps almost entirely, and no AI retrieval system rewards a site that ordinary search cannot read either. Spending on AI visibility while the fundamentals are broken is paying twice for the same fix.
Nobody in your category is being recommended at all. Neither, yet. Test it in an afternoon for nothing: ask the five or six questions a real buyer would ask, in each of the main assistants, and see whether any vendor is named. If the answers are vague and nobody is recommended, the market has not formed and there is nothing to optimise for. Check again in a quarter, because a position that is cheap to establish now gets expensive to buy later.
The question that catches people out
Several agencies in this category now run their own tracking: one has built an AI visibility module into its platform, another built a monitoring tool in early 2024, a third runs its own insights platform. That is a real capability and a good sign about how seriously the firm takes measurement.
It also means the measurement belongs to them. Ask what happens to the history when the contract ends, whether you can see the raw prompt set and change it, and whether the reported figure can be reproduced in a tool you control. A number you cannot audit and cannot keep is a reporting line, not a metric.
The safest arrangement is the dull one: you own a tracker, the agency works against it, and both of you are looking at the same screen.
Frequently asked questions
Can an AI visibility tool improve my visibility on its own?
No. A tracker measures where you appear in AI answers across a prompt set you define. It does not change your content, your markup, your technical setup or what third party sites say about you, and those are the things that move the number. The value of the tool is diagnostic and it is real: without it you cannot tell whether anything you did worked, and you cannot brief an agency on a problem you have not measured. But the tool and the work are separate purchases, and some vendors blur that line in their marketing.
Which should I buy first, the tool or the agency?
The tool, in nearly every case. Entry pricing runs from roughly $29 to $250 a month against agency retainers starting at $5,000, so the measurement costs less than a day of agency time and it is what makes the agency conversation specific. The exception is a company with no internal capacity and a clear commercial problem, where a fixed price audit from an agency is the better first purchase because it tells you why you are absent rather than only that you are.
Do I need an llms.txt file?
For Google, its own documentation says no: it states that you do not need to create new machine readable files or AI text files to appear in AI Overviews or AI Mode. Other systems have not said the same thing, and the file is cheap to publish, so the honest position is that it is low cost and unproven rather than either essential or useless. What it should not be is a line item of any significance in a proposal. If a quote treats llms.txt as a major deliverable, ask what the rest of the engagement contains.
I blocked GPTBot. Is that why I am not showing up in ChatGPT?
Probably not. OpenAI runs separate crawlers for separate purposes: GPTBot collects content for training generative models, while OAI-SearchBot is the one behind ChatGPT's search feature. Blocking the training crawler, which many publishers did deliberately, is a different decision from blocking search. Check which user agents your robots.txt actually disallows before assuming the two are connected, because the fix is a one line change if it turns out you blocked the wrong one.
Can my existing SEO agency do this?
Often yes, and it is worth asking before you start a search. Most of the underlying work is shared: entity clarity, crawler access, structured data, content that answers a question directly, and third party corroboration. The question is whether they are measuring the AI side or assuming it follows from the rest. Ask them what they track AI visibility with, what the number was for a comparable client at ninety days, and whether you can see the prompt set. If they can show you a baseline, you probably do not need to move.
How long before AI visibility work shows a result?
Technical and structural fixes can show up within weeks, because the systems recrawl and re-retrieve continuously rather than on a fixed index cycle. Content and third party work takes a quarter or more, for the same reason it does in conventional search: new material has to be published, found and corroborated. Set the review point at ninety days and expect the early movement to come from the technical side. Anyone promising a visibility result in the first month is describing a fix you could have made yourself.