Buyer’s guide
How to check an AI visibility tool before you buy it
Most tools in this category grade your homework. Far fewer check whether anyone read it. These are four tests you can run yourself, on any vendor including this one, in about ten minutes.
Hiring an agency rather than buying a tool? Four questions to ask an AEO agency before you sign.
We have deliberately not named anyone here. Every specific below came from reading real products, but each artifact is one deploy away from changing, and a page accusing a named company of something they fixed last week is just wrong. The method outlasts the finding, so you get the method.
Does it ask the engines, or grade your markup?
Is this measuring an outcome, or a proxy for one?
- Run the tool and time it. A result that returns instantly did not query an AI engine, because querying one takes seconds.
- Look for a per-engine breakdown in the OUTPUT, not just engine logos on the marketing page. Logos are decoration; a result that names what each engine said is measurement.
- Read the score's explanation. If a score called AI Visibility resolves to something like schemas missing, it is counting your markup, not citations.
- Search their own copy for the phrase on-site. Tools that grade markup usually say so accurately somewhere, because it is true.
The tell
Engine names offered as a dropdown option are a parameter, not a measurement.
Grading markup is a legitimate thing to sell. Selling it as a measure of whether AI engines cite you is a different claim, and the whole category leans on the assumption that the first predicts the second without publishing the correlation.
Who actually built the tool?
Am I buying from the people who made the instrument?
- View source and read the CSS custom property names. Design tokens are usually namespaced to whoever wrote them, so a prefix that is not the brand you are buying from is a strong signal.
- Check the social share image filename. Assets named for the white-label product rather than the brand are common and rarely cleaned up.
- Look at where images and logos are served from. A shared bucket with a tenant identifier in the path means the site is one tenant of a platform, not the platform.
- Search the vendor name plus white label. Platforms that resell tell agencies so on their own marketing, because that is who they are selling to.
The tell
A brand on the page and a different brand throughout the code usually means a skin over a licensed platform.
White-labelling is legal and often sensible. But it changes what you are buying: a reseller cannot fix the measurement, cannot ship a differentiating feature, and cannot tell you how the instrument works because they did not build it. If the number is ever wrong, nobody you can call owns it.
Who profits from the fixes it recommends?
Is the same company grading the work and selling the work?
- Read the upsell after the scan. If the paid tier is a done-for-you fix service, the scan's job is to generate a list of billable work.
- Compare the fix list to their service list. Recommendations that map exactly onto what they sell are a sales funnel wearing a diagnosis.
- Check whether the score would go down if you did nothing and they did nothing. A score that only moves when you buy something is not tracking your visibility.
The tell
A scanner that grades your schema and citations, sold by a firm that sells schema work, directory listings and link building.
This is not an accusation of bad faith. It is a structural conflict: when the instrument and the remedy have the same owner, there is no version of the scan that recommends doing nothing, and that is exactly the recommendation an honest instrument should sometimes produce.
Can you see a real result before creating an account?
Am I being shown a measurement, or a signup form?
- Open any page advertised as a free or instant checker and count the input fields. If the only field is an email address, the tool is a lead magnet.
- Look for a URL box and a run button on the public page. Their absence is the answer.
- If it does require signup, check whether that account is instant or needs manual approval.
The tell
Pages titled free and instant that contain a newsletter box and nothing else.
A vendor confident in the result shows you one. Gating it is a choice about conversion, and it is worth knowing you made a purchasing decision without having seen the product work.
A fifth check: can AI even read their numbers?
Here is a strange thing we keep finding: AI-visibility vendors whose own homepage statistics are injected by JavaScript, which means the proof numbers they sell with render as blank or zero to the very AI engines they claim to get you cited in. You do not have to take our word for it, and we are not going to name anyone from memory. Paste any vendor's homepage, ours included, into the free server-side rendering checker and read what a non-browser client actually receives.
Three more checks on any scan or claim
- Is the free scan dated and sourced? Look for the day it ran and where each number came from. A result with no date cannot be compared with anything later, and you cannot tell whether it was run for you.
- Does a claimed window fit the vendor's age? Look up when the domain was registered. A 90-day results story needs a vendor that has existed for more than 90 days, or an explanation of where the story came from. A rebrand can be a fair one.
- Does the engine count match the product? Count the engines named on the homepage, on the pricing page, and in a sample report. If the numbers differ, ask which engines your plan measures and which are only mentioned.
If the tool makes changes for you
Some tools now publish changes to your site after you click approve. That can be worth paying for, and it raises questions a read-only tool never has to answer. Get these answers in writing before you connect one.
- What exactly does it write on your platform? Titles and descriptions only, whole articles, redirects, structured data? The answer often differs by platform, and the consent screen can say less than the homepage.
- What can be undone, action by action? Ask for a table: each kind of change, whether it can be rolled back, and by whom. "Where supported" is not a table.
- What happens when the read-back fails? A tool that writes should read the live page afterwards. Ask what happens when the page does not match: a retry, a rollback, or a person, and whether approving again can publish twice.
- Who on your account may approve? If every invited user can send a change live, so can the newest person you added.
- How many changes does one click cover? One approval can mean one title or hundreds of product pages. Ask for the largest batch a single click can send live.
- Who grades whether it worked? If the company that made the change also picks the questions and counts the result, the grade has no outside check.
- What is held out? Without questions or pages left untouched, a rise that lifted everything gets credited to the work.
The same 7 questions, answered for us
We do not write to your site. Our agents draft, and a person on your side ships. Here is what that means for each question.
- 1. Nothing. We hold no write access to any site: no plugin, no CMS connection, no publish step. On paid plans the AI CMO drafts a page, and you edit it and put it live yourself.
- 2. Every change on your site is one you made with your own tools, so you undo it the way you undo anything else there. There is nothing of ours to roll back.
- 3. When you mark a draft shipped and give its address, we read the live page once before recording anything. If it does not load, or still carries a placeholder from the draft, nothing is recorded and the message says why. Later, Read the page again shows whether the shipped sentences are still there.
- 4. Owners, admins, managers and editors can approve a draft; viewers cannot. Approving publishes nothing: it marks the draft ready for you to put live.
- 5. One draft per click, for approving and for marking it shipped, and neither click puts anything live.
- 6. The same measurement that runs for every customer. Where the page came from one of our drafts, we are grading our own draft, so the grade compares the questions it targeted with ones you did not change.
- 7. When you mark a page shipped, the tracked questions it did not target become the comparison, chosen before any later measurement runs. If one has already run since the ship date, we refuse to choose them after the fact and say so.
Run the same four checks on us
A guide that exempts its author is an advertisement. So, in the same order:
- 1. Both, and we say which is which. The free audit grades your pages: it asks no engine anything, which is why it is quick, and it says so on the page. After sign-up we query Perplexity, Google AI Mode and AI Overviews, ChatGPT and Claude and store what came back; every figure carries its sample size, and where we have not measured something it says NOT MEASURED rather than showing a zero.
- 2. We built it. The measurement, the scoring and the storage are ours, and every threshold we use is published with the file that enforces it.
- 3. The measurement never grades its own homework. We do sell an implementation package, and we say so out loud: anything we build for you is re-measured by the exact same system as everyone else, with no thumb on the scale. What we still refuse: link building, directory submissions, content packages, and any fix whose success we would get to score more kindly than yours.
- 4. The free audit runs with no account, no card and no email. You see a real measured result before we know who you are.