No, nobody can guarantee that ChatGPT recommends your business, and both OpenAI and Google publish statements pointing the same way about their own products. OpenAI's help pages state that "Placement is not guaranteed". Google's guidance on hiring an SEO states that "No one can guarantee a #1 ranking on Google" and tells you to beware of anyone who says otherwise. A provider can commit to work, to a method and to a measurement. The result belongs to somebody else.
The short version: a guaranteed ChatGPT recommendation, a guaranteed citation or guaranteed AI visibility is not a thing that exists to be sold, whether it is called GEO, AEO or AI SEO. If a proposal in front of you promises the outcome, the four questions further down will tell you what you are actually buying. The difference between a deliverable and an outcome is the whole of this article. What can be sold, and checked line by line, is a list of deliverables, a documented way of measuring, and a report with the raw log attached.
AI Visibility is owned by Lunasol.
Can anyone guarantee a ChatGPT recommendation? No, and here is who says so
The answer is no, and it does not rest on our opinion. It rests on four published statements, one from OpenAI and three from Google.
OpenAI, on how ChatGPT search behaves. Its help pages say that "ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information" and then, in four words, that "Placement is not guaranteed". The same page adds that "Search results and citations can be incomplete, outdated, or incorrect" (OpenAI Help Center).
Google, on guarantees in general. Its guidance for businesses hiring search help puts it plainly: "No one can guarantee a #1 ranking on Google." It goes further and tells you what to treat as a warning sign: "Beware of SEOs that claim to guarantee rankings, allege a 'special relationship' with Google, or advertise a 'priority submit' to Google" (Google Search Central).
Google, on paying for position. "Google never accepts money to include or rank sites in our search results, and it costs nothing to appear in our organic search results" (Google Search Central).
Google, on its AI features specifically. "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary", and you do not need to "create new machine readable files, AI text files, or markup to appear in these features" (Google Search Central).
Read those together and the position is clear enough for a buyer. Neither company's documentation describes a submission form, a partner programme or a payment that puts a business into an answer, and Google states plainly that it never accepts money to include or rank sites. OpenAI's help page does not address payment at all; it says only that placement is not guaranteed.
Where a provider tells you a route in exists, ask them to show you the published page that says so, and check it against the four statements above. Gathering those took about five minutes.
How does ChatGPT choose which businesses to recommend? What you can and cannot control
On the pages we checked, neither OpenAI nor Google publishes how an answer picks one business over another. What the documentation does name is two switches you control, and nothing that puts you into an answer: whether a search crawler may fetch your pages, and whether your content may be used in training. Both are below, with what each one does not do. What your pages say and whether your facts agree across the places they appear are not in the documentation either; they are ordinary business hygiene, and they are the work a proposal should be putting a price on.
- You control whether a search crawler may fetch your pages. OpenAI documents that "OAI-SearchBot is used to surface websites in search results in ChatGPT's search features", and that "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links" (OpenAI). That is a switch you hold, and it only works in one direction: you can remove yourself, not add yourself.
- Training access is a separate switch, and it is not the search one. "GPTBot is used to make our generative AI foundation models more useful and safe", and "Disallowing GPTBot indicates a site's content should not be used in training generative AI foundation models." Two different decisions, and a business is allowed to make them differently. Our article on whether AI search can access your website works through each of these crawler names one at a time.
- A user-triggered fetch decides nothing about search. "ChatGPT-User is not used to determine whether content may appear in Search." So somebody pasting your link into a chat is not a route in.
- Advertising is documented separately from search, and the documentation does not connect the two. OpenAI documents a crawler whose job is to "validate the safety of web pages submitted as ads on ChatGPT" (OpenAI). That an advertising product exists is not published evidence either way about an organic answer, so if it is offered to you as proof that placement can be bought, ask for the line on the page that says so.
- On Google's side, there is no extra layer to buy. No special markup, no AI file, no additional requirement. The best practices for search "remain relevant for AI features in Google Search" (Google Search Central), which is a way of saying there is no separate game.
So the honest list of what you control is short, and the answer itself is not on it: no amount of money moves it there. What is on it is whether you can be reached, what your pages say, whether your facts agree with each other across the places they appear, and how often you check. In this market, the places your facts appear usually come down to your business profile, the directory your sector uses here, and whichever of your pages states your emirate and the name on your licence. Every item on that list is work somebody does on a schedule, which is why a process worth buying is laid out as a sequence of steps; Lunasol publishes its own in four (Lunasol).
What can a GEO agency actually guarantee? Deliverables against outcomes
A deliverable is something the provider does and can hand you afterwards; an outcome is something a third party does in response, and no provider controls it. That line settles most arguments about proposals.
A deliverable exists whether or not it works. That is the awkward half of the line, and also the half you can check.
Outcomes nobody can promise, against the deliverables a provider can date, hand over and be checked on
| Outcomes, which cannot be promised by anyone | Deliverables, which can be promised, dated and checked |
|---|---|
| Being recommended by an assistant | A written prompt set, and the runs against it |
| Being cited in an answer | A dated baseline you can compare against |
| Appearing above a competitor | A named competitor set, fixed before the work starts |
| A position, a rank or a slot | A list of what is wrong on your site, in priority order |
| A number of mentions per month | A number of pages written or rewritten per month |
| Traffic from an assistant | Changes made to your business profile and directory records |
| A share of voice figure | A report on a stated schedule, with the raw log attached |
| Anything expressed as "we will get you..." | Anything expressed as "we will do, by this date, and show you" |
Read each column as its own list. The rows are not matched pairs.
The test for any line in a proposal: could the provider deliver it entirely on their own, without anybody else agreeing? If yes, it is a deliverable, and it is the kind of line you can ask to see evidence of. If it needs a third party to cooperate, it is an outcome, and no provider controls it. What either line means inside your agreement is a question for your own adviser, not for us.
That does not make outcomes unimportant. They are the reason you are buying. It makes them the thing you measure rather than the thing you contract for, which is why the measurement has to be defined before the work starts rather than argued about afterwards. Our article on how to check what AI says about your company sets out how to run and record that measurement so it is comparable the second time.
Four questions to ask before you sign an AI visibility guarantee
Four questions expose a guarantee claim: what exactly is guaranteed in a noun you can count, what triggers the remedy, which part depends on somebody outside the room, and whether you can see an anonymised example of every deliverable. Ask them before you sign anything. Each has an honest answer, and the honest answer is usually more useful than the promise it replaces.
1. What exactly is guaranteed, in one sentence, using a noun I can count?
An honest answer names a deliverable: a number of prompts, a number of runs, a number of pages, a report date. An answer that reaches for visibility, presence, prominence or exposure without a unit attached is not describing anything that can be checked.
2. What happens if it does not occur, and what triggers that?
Expect something specific about the trigger and the remedy, in writing, and usually narrower than the headline suggested. Ask to see the clause rather than hear a summary of it. A remedy of continued work at no charge is worth knowing about, but it is not a guarantee, and a provider who says so plainly is being straight with you.
3. Which part of this depends on somebody outside this room?
All of the outcome does. Providers who work this way tend to volunteer it early, because it is how they scope. If they tell you that no part of it depends on anybody outside the room, that does not match the documentation quoted at the top of this article. Ask which published page their answer is based on.
4. Show me an anonymised example of every deliverable you have just listed.
This is the quickest of the four to ask and the easiest to answer. A provider who has done the work has a redacted prompt set, a redacted run log and a redacted report, and can show you all three in a call. A description of a deliverable is not a deliverable.
If the four answers are specific, you are probably talking to somebody who has done this before, whatever they call themselves. If they are not, the price is irrelevant.
What can an AI visibility agency honestly promise?
Nobody can promise you a recommendation, a citation or a position in any AI product. What can be promised is quite a lot, and all of it is checkable: the prompt set, the runs, the baseline, the fix list, the pages, the corrections and the report, on dates you agree in advance, with the log attached so you can check the work rather than take it on trust.
What a provider can commit to, what you should be handed, and how to check it
| The commitment | What you are handed | How you check it |
|---|---|---|
| A written prompt set | The file of questions, before the work starts | Open it and read the questions. They should be questions your customers ask, not questions with your brand name already in them |
| The platforms in scope | A named list, and a named list of what is out of scope | Count them, and check the second list exists at all |
| A stated number of runs per question | The environment recorded against each run | Look for the date, the account type and the location on every row |
| A competitor set fixed in advance | The list, agreed before any work | Check it is the list you would have written, and that it cannot be edited later |
| A dated baseline | The first measurement, before anything changes | Check the date is earlier than the date work started |
| A list of what is wrong and missing on your site | The list, in priority order, with an owner per item | Pick three items at random and see whether you can act on them as written |
| A stated number of pages written or rewritten | The pages, on the dates agreed | Read one and see whether it answers a question or describes a company |
| Corrections submitted where your facts are wrong | The submissions, logged with dates | Acceptance is somebody else's decision, so check the log of what was sent, not what was accepted |
| A report on a stated schedule | The report, with the raw log attached | Open the log. A summary without a log cannot be audited |
| A re-test date | The date, agreed in writing before the work starts | Check the date is in the agreement rather than in an email, and that the questions named for it are the same ones in the prompt set |
| A written list of exclusions | The list | If it does not exist, the scope is whatever anybody later says it was |
Every line in that table can be shown to you, dated and compared against what was promised, and none of it needs a third party to agree to anything.
Why do agencies sell guaranteed AI visibility?
Agencies sell guaranteed AI visibility for four structural reasons: buyers ask for a guarantee, the vocabulary is loose, the work is invisible, and the promise is cheap to make and hard to enforce. None of those reasons is about anybody's motives, and knowing them makes the pitch easier to read.
Buyers ask for it. A guarantee is the natural thing to want when you cannot evaluate the work. If two proposals look identical and one promises an outcome, the promise does the choosing for you. That is not a trick played on you. It is a gap in what you can assess, and the fix is a better test, not a better promise.
The category is young and the language is loose. Words like visibility, presence, ranking and mention are used interchangeably in proposals, and some of them are measurable while others are not. Ask which unit the noun in your proposal is counted in.
The work is invisible. Most of what a competent provider does happens in places you never look: your own page text, your business profile, your structured data, directory records. Nobody buys a directory record they will never see, so something they can see goes on the front of the proposal.
A guarantee is cheap to make and hard to enforce. Ask yourself what actually happens if it is not met. If the answer is a refund, ask for the refund terms in writing and read what triggers them. If the answer is more work at no charge, that is not a guarantee, it is a retainer with a different name on it.
None of this means a provider promising results is acting in bad faith. It means the promise cannot be kept by anybody, so it tells you nothing about which provider is better.
We asked an AI assistant how ChatGPT chooses businesses: what it said
In September 2026 we put two questions to Google's AI Mode: whether an agency can guarantee a ChatGPT recommendation, and how ChatGPT decides which businesses to recommend. One run each. We also tried a second assistant, which would not send the prompt at all, so this is one tool on one day and not a study.
On the guarantee question, it agreed with the documentation. Its first line was "No, an agency cannot guarantee that ChatGPT will recommend your business." It wrote its own heading, "Why Guarantees are Impossible", said the product does not "offer a paid advertisement section like Google Ads where you can buy your way to the top", and concluded that "no third party can force a recommendation". It then warned the reader directly: "If an agency offers a zero-risk 'guarantee' to list you on ChatGPT, it is a major warning sign. Look for partners who talk about improving visibility signals rather than locking in permanent rankings."
On the mechanism question, it produced a great deal more than anybody has published. It set out five numbered pillars with sub-points, attached a percentage range to the share of citations coming from one kind of page, and said that "data suggests" a smaller number of consistently positive reviews outperforms a larger number of mixed ones. We have not verified any of that and we are not repeating it as fact.
Neither answer cited OpenAI. Across the two answers there were forty-nine outbound links, twenty-two in the first and twenty-seven in the second. Each answer drew on nine distinct external hosts. Not one of those hosts was OpenAI. The visible source cards were agency blog posts and a video in the first answer, and a discussion-forum thread and two agency blog posts in the second. The first answer reached the same conclusion as OpenAI's published wording, but it did not get it from there. Nothing in the answer lets a reader tell the difference.
One of the sources it drew on says the mechanism is not known. The preview text on one of the cards the second answer drew from began: "A necessary caveat. Nobody outside OpenAI knows the exact ranking mechan...". The answer built a five-pillar mechanism on top of a page that opens by saying the mechanism is not known outside the company.
That is why an assistant's description of itself is a self-report rather than evidence. It is assembled from pages about the product, and in this check those pages were agency blog posts, a video and a discussion thread rather than the platform's own documentation. When you want to know what a platform does, read the platform.
Where Lunasol stands on this
Lunasol, which owns AI Visibility, publishes a four-step process: an audit it calls "The AI census", signals work to "Build what AI trusts", content described as "Write the way AI reads", and quarterly tracking, "Numbers, every quarter" (Lunasol). The tracking step is described on that page as re-asking "the models your customers' questions", which is a measurement on a schedule: a question set put again, with what it finds written down. What an audit of that kind covers is set out in our article on what an AI search audit includes.
Run those four names through the test above. Each one names work rather than a result: an audit, a set of changes, a quantity of content, a measurement on a schedule. A step name is not a scope, though, and a published process is not a contract. What any provider will actually hand you, and on what dates, is set out in the proposal they give you rather than on their website, and that goes for Lunasol too. Ask for the proposal.
Where to start with the proposal you already have
Take the proposal you already have and mark every line as either a deliverable or an outcome. The test is the one above: could the provider deliver it entirely on their own, without anybody else agreeing? Then ask the four questions. That is twenty minutes, and it will tell you more than another round of quotes.
If you would rather see where you stand first, Lunasol, which owns this publication, offers a free AI visibility check as a starting point. Its published process and its contact routes, WhatsApp and email, are at Lunasol: there is no form, and the stated reply time is within the hour.
Keep the marked-up proposal either way. It is the document you will compare the next one against.
Common questions
Can anyone guarantee that ChatGPT will recommend my business? No. OpenAI's own help pages state that "Placement is not guaranteed", and Google's guidance on hiring search help states that "No one can guarantee a #1 ranking on Google." A provider can guarantee deliverables, dates and a report. Nobody can guarantee the answer.
Can I pay to appear in ChatGPT or in Google's AI answers? Not for an organic recommendation on Google's side: Google states that it "never accepts money to include or rank sites in our search results, and it costs nothing to appear in our organic search results" (Google Search Central). OpenAI's help page says nothing either way about payment, and the only thing it documents in that direction is making your site crawlable, which makes you eligible to be surfaced and nothing more. Advertising products exist, and no published page we checked states that buying one affects an organic recommendation. Treat anyone who tells you it does as making a claim they need to show you in writing.
Is there a form to submit my business to ChatGPT? None is described in the documentation we checked. What that documentation describes is the opposite direction: a crawler that can be allowed or disallowed. Allowing it makes you eligible to be surfaced; it does not put you anywhere.
How does ChatGPT choose which businesses to recommend? On the pages we checked, neither OpenAI nor Google publishes the mechanism. OpenAI states only that "ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information" and that "Placement is not guaranteed". Anybody quoting you a numbered mechanism is quoting something neither company has published.
Are AI visibility guarantees different for a UAE business? The platform documentation quoted here is the same documentation wherever you are based, and none of it is written for one country. We say nothing about how any UAE rule treats a guarantee or a contract; that is a question for your own adviser. What is local is which sources carry your facts: your business profile, the directories your sector uses here, and whichever of your pages states your emirate and the name on your licence.
How long before I know whether the work has done anything? Ask for the re-test date in writing before you start, and for the same questions to be asked the same way on it. A quarter is a reasonable gap, and the one we would ask for. A provider who will not name a re-test date has not defined the measurement.
What if an agency shows me a screenshot of my competitor being recommended? A screenshot shows one answer, on one account, at one moment. It is evidence that the answer happened, not evidence that anybody caused it or can repeat it. Ask how many runs it came from and what the other runs said.
Is GEO a scam, or is this work worth paying for? Not a scam as a category. The work and the guarantee are different things, and it is the guarantee that nobody can deliver. Whether to buy the work is a commercial decision, and it depends on the deliverables, not on the promise. The work is the same work a careful business would do anyway: state your facts clearly, make them consistent everywhere they appear, make sure your pages can be read, and check what is being said about you on a schedule.
What should I do if I have already signed a contract with a guarantee in it? Read what the guarantee is attached to and what triggers the remedy, and take the contract to your own adviser. We are a publication, not your lawyer, and nothing here is advice about your agreement.
What nobody can tell you about how an AI answer is built, including us
How any particular answer is assembled is not on the pages we checked, from either OpenAI or Google, and no provider quoting a mechanism at you is working from a published one. Nor are we. The documentation describes controls and eligibility. It does not describe a formula, and the confident mechanism accounts circulating online are published by neither company; in the answers we logged, the pages behind them were third-party blog posts, a video and a discussion thread rather than vendor documentation.
None of that is a reason for despair. It is that the work has to be justified by the work: is your information correct, is it reachable, is it consistent, and are you checking. Those are worth doing whether or not any particular answer ever mentions you, and none of them requires anybody to promise you anything.
Method, sources and notes
This article is general information about buying and scoping a service, current at September 2026. It is not legal, contractual, advertising, consumer-protection or purchasing advice, and nothing in it describes what any UAE authority requires or permits, or what any contract or promotional claim may lawfully say. If a proposal or a signed agreement in front of you is unclear, take it to your own adviser. No provider, of any kind and by any method, can secure a mention, a citation, a ranking or a position in any AI product, and nobody can promise one.
Nothing here is an accusation against any provider, named or unnamed. We name no company and no domain from the live check, and we describe no provider's conduct. The structural reasons given for why guarantees get sold are offered as explanation rather than as a finding about anybody. Nothing here is a recommendation of, a comparison between or a ranking of any supplier, and we neither endorse nor criticise any firm, named or unnamed. The article sets out a general test rather than an assessment of any particular provider, and it makes no statement about what any proposal, quotation or agreement from anybody says.
The live check was one tool, one day, one connection, two questions and one run each, which is an illustration of a method and not a study. The protocol planned four answers, two prompts in each of two tools, and produced two. On the second assistant only the first prompt was ever attempted, once, and it could not be sent; the second prompt was never put to it, and since we established no cause we neither name the tool nor attribute the failure. Repeat-run variability was out of scope here and we did not test it. We report counts, structure and verbatim sentences about guarantees. Where we quote what an assistant said about how a system works, we quote it as a self-report and we verified none of it; in particular the figures that appeared in the second answer are reproduced nowhere in this article as fact. Where a quotation we reproduce contains quotation marks of its own, we render those inner marks as single marks; no word or order is changed. The logged runs are available on request. Assistant answers vary by account, location and time, and quoted product documentation changes without notice, so check the current version of any page quoted here before relying on it.
Product and company names are used here for identification only. Neither this publication nor its owner is affiliated with, endorsed by or sponsored by any of them, nothing here is an endorsement by us of any of them, and none of them has reviewed, approved or contributed to this article. ChatGPT, OAI-SearchBot, GPTBot and ChatGPT-User are products of OpenAI, and the OpenAI Help Center is a publication of OpenAI. Google Search, Google Search Central, AI Mode and AI Overviews are products or publications of Google. Google Ads is named only inside a quotation of an assistant's answer and has no connection with this article. WhatsApp is a product of Meta. All are trademarks of their respective owners.
AI Visibility is owned by Lunasol.
