GEO Pricing in Dubai: What Determines the Scope of Work?

You are holding two or three proposals for the same job and the totals are nowhere near each other. Here is why. GEO, generative engine optimisation, is work aimed at getting your business named and described accurately when a customer asks an AI assistant a question instead of typing it into a search box. An answer surface is one place an AI answer appears, such as ChatGPT, or Google's AI Overviews, or Google's AI Mode, and each one has to be checked separately. Seven things move the size of that job, and none of them is the provider's opinion of your business: how many questions get tracked, how many competitors you are scored against, how many answer surfaces are checked, how many languages, how many page templates get changed, whether anybody implements the changes or only advises, and how often it all repeats. Read those seven off a proposal and you can tell why one quote is several times another without either being dishonest. This article defines each of them, says which are about your business rather than the provider's rate, and sets out what no scope of any size can buy.

The short version: a quote is a scope, not a price. Before you compare two totals, compare the seven variables behind them. A cheaper proposal is doing less, doing the same work more efficiently, or doing it badly, and the total on its own will not tell you which. The seven variables separate the first of those from the other two, by showing you how much work is actually in each proposal. Telling efficient from bad is a second test, and the questions for it are further down, taken from Google rather than from us. This article carries no price, ours or anybody's, and the evidence section explains why: not one of the figures that came back in our own live check had anybody neutral behind it. What travels between businesses is the scope, so that is what this article gives you. The only reliable way to turn it into a number is this: fill in the four your own business already decides and send the same four to every provider, then ask each of them to write down the assumption they are making on the other three. Put those assumptions side by side, settle on one set, and have everybody price that. Quotes built on the same quantities can be compared. Totals on their own cannot.

AI Visibility is owned by Lunasol.

Why do two GEO quotes for the same business differ several times over?

A GEO proposal is a scope of work with a number at the bottom. The number is arithmetic on the scope. So two honest providers looking at the same business can land a long way apart, because they have assumed different answers to the same seven questions and nobody wrote the assumptions down.

That is the common case, and it is not anybody behaving badly. One provider assumed forty tracked questions and one platform; the other assumed four hundred and four platforms. One assumed you would implement the changes yourself; the other assumed they would. One priced a quarterly re-measurement; the other priced a single report. Those are four different jobs with one name on them.

There is a separate thing to check for, and it is not the same as a proposal being cheap. A proposal that does not name its numbers is not cheaper than one that does, it is unscoped, and an unscoped proposal cannot be compared with anything. That is a gap to ask about rather than a verdict to reach, because most proposals have the numbers and have simply not written them down. What is being promised, as opposed to how much of it, is a different question again.

What makes a GEO project bigger or smaller? The seven variables

The seven variables that move the size of a GEO engagement are the questions tracked, the competitor set, the answer surfaces, the languages, the page templates in scope, implementation and repetition. Every proposal you are holding has an answer to all seven, whether or not it has written them down. Each one below comes with what makes it bigger and what makes it smaller.

Questions tracked. This is the biggest single driver and the one most often left vague. A prompt set is the list of questions the check is run against, and it is written down or it does not exist. Our article on how to check what AI says about your company covers how to build and record one. The number matters twice over, because every question is asked on every surface, in every language, against every competitor, every time the check repeats. It is the variable that multiplies all the others.

Competitor set. The competitor set is the named list of businesses you get scored against, and it has to be fixed in advance, or the second measurement is not comparable with the first. That first measurement is your baseline, the reading every later one gets compared to, which is why changing the list afterwards costs money. A longer list is more work in a straight line. A list nobody can agree on is more work than either, because it gets revised halfway through and the baseline is lost.

Answer surfaces. Each assistant is a separate check with separate behaviour. The temptation is to buy all of them. The better question is which ones your customers actually use, and that is a question about your buyers that you can answer and nobody else can. Note also that surfaces differ in what they will even show you: Google states that sites appearing in AI features "are included in the overall search traffic in Search Console. In particular, they're reported on in the Performance report, within the "Web" search type" (Google Search Central), so some of what you are paying to watch is watched by asking rather than by reading a report.

Languages. A second language is not a translation line on an invoice. Part of the cost is a piece of plumbing called hreflang. It is an annotation on each language version, usually a tag in the page itself, and it tells Google this page has an Arabic twin and here it is. Google states plainly that it "doesn't use hreflang or the HTML lang attribute to detect the language of a page; instead, we use algorithms to determine the language", and that for the annotations to work at all "Each language version must list itself as well as all other language versions", because "If two pages don't both point to each other, the tags will be ignored" (Google Search Central). That is upkeep on every page you annotate, for as long as both versions exist, although the same documentation allows some languages to be left off some pages and says Google "will still process the ones that point to each other". On top of it sit the things our article on English and Arabic AI search works through: the same company name spelled several ways, question phrasings that are not translations of each other, and a separate set of results to record. A bilingual scope is closer to two scopes than to one and a bit.

Page templates in scope. Not pages, templates. Fixing one product page fixes one page; fixing the product template fixes the catalogue. This variable is how much of your site is in scope; the next one is who actually touches it, and the two are priced separately. The count that matters in a proposal is how many distinct templates get touched, plus how many separate platforms they live on. Some of that work is structured data, also called schema: a block of labelled facts added to a page, invisible to your customers, that tells a machine which part of the page is the product name, which is the price and which is the image. You do not need to understand it in order to buy it, but you can count it, because where structured data is in scope the work is bounded by published lists rather than by taste. Google's merchant listing documentation requires name, image and offers on the product, and on the offer a price, a currency and a price specification, with the price and the currency allowed either directly or inside that price specification. It then recommends nineteen further properties on the product and eight on the offer (Google Search Central). A proposal can be asked how many of those it is doing and on how many templates. Our article on the product information a UAE online store should publish works through the full list.

Implementation. The single largest swing between two quotes, and the one most often invisible. Advice is cheaper than work. A proposal that hands you a list of changes and a proposal that makes them are different purchases, and the second one carries a dependency the first does not: access to your site, your platform, your developer's time and your approval cycle. If a quote seems low, check this line first. If it seems high, check whether you are paying somebody to do something your own team was going to do anyway.

Repetition. A baseline measured once is a photograph. The cost of the second and subsequent measurements is what turns it into a measurement, and it is a recurring line rather than a one-off. Note the lag this imposes on any promise about speed: Google says its search data should normally be available "in 2-3 days" and that "To protect user privacy, the Performance report doesn't show all data" (Search Console Help), while attribution in analytics only looks back so far, with a thirty-day default for acquisition key events and ninety for the rest (Google Analytics Help). Our article on measuring what AI search brought in covers the reporting side.

What happened when we asked an AI assistant what GEO costs in Dubai

The first thing most buyers do, before they read a proposal properly, is ask an assistant what GEO should cost in Dubai. So that is what we did, twice, in September 2026, in Google's AI Mode, one run each. What came back is the reason this article carries no figure.

It answered with a price before it asked anything about scope. The first question was about cost in Dubai, and the first sentence of the answer was a monthly range in dirhams. The answer then built a three-tier monthly price table, added a separate one-off fee range for an audit, and finished with a four-column comparison of service models against price ranges. Fifteen dirham figures in one answer.

Every figure came from somebody selling the work. This is the finding that matters. Each tier and each cost factor in that answer carried a source chip naming a commercial supplier of the service, several of them with a further indicator showing more sources of the same kind behind it. No neutral survey, no trade body and no published index was cited anywhere in it. What the answer had done was gather what suppliers publish about their own services and set it out as a table. It reads like a market. It is one side of a market, in a layout that does not say so.

Ask the same thing without the city and the currency changes. The second question dropped Dubai and asked what the price of an AI visibility project depends on. Zero dirham figures. Fourteen dollar figures. Nothing about the work had changed between the two questions. What changed was which pages the answer was built from, and those pages quote in different currencies.

The same item, priced roughly twenty times apart. Both answers put a figure on a one-off audit. Converted into one currency, the two ranges differ by roughly a factor of twenty. Same tool, same day, same connection, same item, two phrasings of one question. We are not reprinting either range. A figure with no authority behind it becomes an anchor the moment a reader sees it, and neither answer had any authority to give one.

Both answers then qualified us. Each one finished by putting three qualifying questions to us, and the first invited us to name our industry for a more specific estimate. That is an intake form. It is a reasonable thing for an assistant to do and a strange thing to mistake for market research.

A guarantee appeared inside a tier. In the first answer, one of the tiers was described as including digital PR that would guarantee brand mentions in AI datasets. Those are the assistant's words in the answer we read. Its source chips pointed at suppliers of the service. We read the chips rather than opening the pages behind them, so we are not reporting that any supplier published that promise. What the wording runs into is documentation: OpenAI's own help pages say of its search product that "Placement is not guaranteed" and that "Search results and citations can be incomplete, outdated, or incorrect" (OpenAI Help Center). A guarantee of mentions is the exact thing our article on whether you can guarantee a ChatGPT recommendation works through, and finding one inside a price tier is a reminder that a tier in a generated table is copy, not a specification.

What the two answers were good at was naming the variables. Between them they named prompt volume, competitor set size, the number of answer surfaces tracked, audit against implementation, content and technical work, industry competition, geography and multi-location complexity, language, and the engagement model. Most of that list is the list above. On the variables, we and the sell side largely agree. The answers also named the competitiveness of your industry and the number of locations you trade from, which this article treats inside the competitor set rather than as lines of their own. It is the numbers, not the variables, that nobody can source. That is the good news: the list a seller uses to build a quote is a list you can hold them to.

Limits. One tool, one day, one connection, two questions, one run each, link counts read rather than clicked. It is an illustration of what a buyer is told when they ask, not a survey of the market, not a measurement of what anybody charges, and not a finding about any supplier. We name no agency, publication, product or domain from either answer.

Which GEO scope variables are about your business, not the provider's rate

Four of the seven variables are settled by your own business before anybody quotes, and you can fill them in today.

• The questions tracked. You already know how many services you sell and how many kinds of buyer you have, which sets the size of the question list.
• The languages. You already know whether you operate in one language or two.
• The page templates in scope. You already know how many templates your site has and whether it is one platform or three.
• Implementation. You already know whether you have somebody who can implement changes or whether you do not.

Work those four out before you read a single proposal and two things happen. You stop comparing totals and start comparing assumptions. And you can ask for a proposal to be re-scoped on a named variable, which is a conversation, rather than asking for a discount, which is not.

The other three, the competitor set, the surfaces and the repetition, are decisions you make with a provider rather than before one. They are still yours. If a proposal treats the competitor set as the provider's call rather than yours, ask to see the list and ask what happens to the baseline if it changes.

What does a bigger GEO scope not buy?

No scope, at any size, buys a mention, a citation or a position in an AI answer. The outcome is the one thing that does not change with the size of the work.

Google's own guidance for people hiring search help states flatly that "No one can guarantee a #1 ranking on Google", and lists among its warning signs SEOs who "claim to guarantee rankings, allege a 'special relationship' with Google, or advertise a 'priority submit' to Google" (Google Search Central). For its own AI surfaces specifically, Google states that "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" and that "You don't need to create new machine readable files, AI text files, or markup to appear in these features" (Google Search Central). OpenAI states that "Placement is not guaranteed" (OpenAI Help Center).

Read those together and the pricing implication is exact. A larger scope buys more work, done more often, across more surfaces and more languages. It does not buy a mention, a citation or a position, and a tier that says it does is describing something nobody has.

There is a second thing that does not change with scope: the idea that AI surfaces need a separate layer built for them. For Google's AI Overviews and AI Mode, the documentation quoted above describes no separate file, no separate markup and no special optimisation required to appear. So if any proposal is priced on unlocking a mechanism a smaller one cannot reach, ask it to put that mechanism in writing and to name the documentation it comes from.

How to compare two proposals that are scoped differently

Do not compare the totals. Copy or screenshot the grid below, put each proposal in a column, and fill in what you can. Every cell is a number or a yes or a no, never a paragraph. Three things will happen.

Expect blanks, because plenty of proposals do not write their numbers down. Blank cells are the finding, not a failure of the exercise. Ask for each one in writing, so the answer becomes part of the proposal rather than part of a conversation.

Some cells will differ by a factor you can name. That is the honest gap, and it is now a decision you can take: fewer questions, fewer surfaces, one language first, implementation in house.

And the totals will stop being comparable in the way you assumed, which is the point. A quote that is half another quote and covers a third of the work is buying less per unit spent, which is worth knowing before you decide whether the missing two thirds is work you wanted.

Questions to ask, from Google rather than from us. Google's own page for businesses hiring search help lists eight questions to ask a prospective SEO, and they apply to a GEO proposal without modification. These are also the test for whether a cheap proposal is efficient or simply poor, which the seven variables on their own cannot tell you. Six of the eight matter most here:

• "Can you show me examples of your previous work and share some success stories?"
• "Do you follow the Google Search Essentials (previously known as Webmaster Guidelines)?"
• "What kind of results do you expect to see, and in what timeframe? How do you measure your success?"
• "What's your experience in my industry?"
• "How long have you been in business?"
• "How can I expect to communicate with you? Will you share with me all the changes you make to my site, and provide detailed information about your recommendations and the reasoning behind them?"

All six are from Google Search Central. The last one is the most useful and the least asked.

An example scope from Lunasol

A deliverable is a named thing you get: an audit, a report, a set of pages changed. Here is the deliverable half of a scope, taken from the deliverables Lunasol publishes on its own page. Every line of it is public, and it is here so you can see the shape of a scope before you put one next to a number. Like any published list it names the work and leaves the seven quantities to a buyer, apart from the tracking rhythm, which it fixes at quarterly. It carries no figure, because this article carries none, ours or anybody's. It is not a comparison either: no other provider is named anywhere in this article, including the suppliers the assistant cited.

Example scope from Lunasol Lunasol, which owns AI Visibility, publishes a four-step process: Audit, which it calls "The AI census"; Signals, "Build what AI trusts"; Content, "Write the way AI reads"; and Track, "Numbers, every quarter" (Lunasol). Its published GEO deliverables are an "AI visibility audit, what ChatGPT says about you today", "Entity and brand signal building", "Content structured for AI citation", "Presence in sources AI models trust and quote", "Schema and structured data implementation", and "Quarterly AI visibility tracking". AEO, answer engine optimisation, is the narrower half of the same job: work aimed at a direct answer to a specific question being shown, rather than at your business being named in general. Its published AEO deliverables are "Question research, what your customers actually ask", "FAQ and answer-block content structure", "Schema markup, FAQ, HowTo, and more", "Featured snippet targeting for money queries", "Optimization for Google AI Overviews", and "Tracking of answer placements won". One word in that list needs translating. An entity is your business treated as a thing a machine can recognise and keep straight: the company with this name, at this address, in this line of work, as distinct from every similarly named business elsewhere. Read against the seven variables: question research sets the question list, entity and signal work and the schema implementation are the template count and the implementation line, quarterly tracking is the repetition, and the audit is the baseline the rest is measured from. Of the three a buyer normally settles with a provider, this page fixes the repetition already, at quarterly. The competitor set and the answer surfaces are what no published page can settle in advance, and a buyer settles those in a first conversation with anybody. These deliverable names are quoted from a published page. A deliverable name is not a promise, on this page or on anybody's. What no scope contains at any size is the outcome itself: the section above on what a bigger GEO scope does not buy applies to this list exactly as it applies to every other provider's. How a number gets built from a scope like this. The deliverables are the work, the seven variables are the quantities, and a quote is those two put together. Fill in the quantities and every provider is pricing the same job, which is the only thing that makes their numbers comparable. That is the rule, and it is the whole of the rule.

On price, plainly. No price appears in this article, ours or anybody's, and that is an editorial decision rather than an oversight. A figure printed in a publication owned by an agency would be read as a market rate within a day of publication, and an unsourced figure anchors a reader whether or not anybody stands behind it. What a buyer can honestly be given is the scope, the variables and the questions. Our article on writing a service page covers the same problem from the seller's side, including the pricing block itself. Ask for the number the way you would ask anybody: against a written scope, with the seven variables filled in. Lunasol's published process and its contact routes are at Lunasol.

Where to start this week

Fill in the four variables your own business already decides before you read another proposal. How many services, how many buyer types, one language or two, how many templates on how many platforms, and who implements. That is an hour with a notepad and it changes every conversation that follows.

Then take the proposals you already have and fill in the grid, and go back to each provider for the cells you cannot fill yourself. Most will fill them once asked, and the grid then turns a guess into a decision.

If you would rather have the scope worked out with you, the starting point Lunasol publishes is an "AI visibility audit, what ChatGPT says about you today", which is the baseline the seven variables get filled in against. Send the four your own business already decides and ask for the other three to be scoped with you, which is exactly what this article tells you to ask of anybody. Its published process and its contact routes are at Lunasol: WhatsApp or email, no form to fill in, and the page states replies within the hour.

Common questions

How much does GEO cost in Dubai? The cost of a GEO engagement in Dubai depends on its scope, and no figure means anything until that scope is written down, because the total is arithmetic on seven variables that two providers can answer very differently. This article deliberately carries no figure. When we asked an assistant the same question it produced a tidy table of monthly ranges, and every figure in it came from a company selling the service. Work out your seven variables first, and the number follows from them.

Why is one GEO quote several times another? Two GEO quotes for the same business usually differ because each has assumed different numbers for the same seven variables: how many questions are tracked, against how many competitors, on how many answer surfaces, in how many languages, across how many page templates, with or without implementation, and how often it repeats. Two proposals can be honest and still differ several times over.

Is a cheaper GEO proposal worse? Not necessarily. A cheaper proposal is doing less, doing the same thing more efficiently, or doing it badly. The seven variables tell you whether it is doing less. If the quantities match and the totals still do not, that is when you ask the hiring questions further up, which are Google's rather than ours. A proposal that will not state its numbers is not cheap, it is unscoped.

What should a GEO proposal include? A GEO proposal should state a number for each of the seven variables: how many questions it tracks, how many competitors it scores you against, how many answer surfaces it checks, how many languages, how many page templates it changes, whether it implements or only advises, and how often it repeats. A proposal that names none of those cannot be compared with anything, which is a gap to ask about rather than a verdict on the provider.

Does a bigger budget get me into AI answers? No. Google states that there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations necessary, and that no one can guarantee a number one ranking on Google. OpenAI states that placement is not guaranteed. A larger scope buys more work, more often, across more surfaces. It does not buy an appearance, and nobody can promise one.

Should I pay for a one-off audit or a monthly retainer? An audit and a retainer answer different questions. An audit tells you where you stand once, and our article on what an AI search audit includes sets out what belongs in that report. A retainer is a recurring monthly fee for continuing work rather than a one-off project fee, and here it buys the second measurement and the third, which is what makes the first one comparable with anything. If you buy only the audit, buy it knowing you have bought a photograph.

How many AI surfaces should I pay to track? As many as your customers use, which is a question about your buyers rather than about what is on offer. Each surface is a separate check, so this is a variable that multiplies rather than adds.

Does Arabic really double the work? A second language doubles whatever gets done once per language, and it adds maintenance that does not exist in one language at all, because the language annotations have to point at each other or Google ignores them.

Why does this article carry no price? Because this publication is owned by an agency that sells this work, and a number printed here would be read as a market rate rather than as one company's quote against one scope. The variables travel; a price does not. Work out your seven, then ask for a number against them.

What this article cannot tell you

This article cannot tell you what anything costs, and it has not tried. It cannot tell you whether a particular proposal is fair, because fairness is scope against price and we can only see one of those. It cannot tell you what any other provider charges, and any figure you find online, including the ones an assistant hands you, is somebody's marketing until you can name whose.

What it can do is stop the two mistakes that cost the most. Buying on a total, which means buying an unknown quantity of work. And assuming the expensive proposal must be doing more, when the only honest way to find out is to make both of them write their numbers down.

Method, sources and notes

This article is general guidance on scoping and comparing marketing services, current at September 2026. It is not legal, tax, accounting, procurement or contract advice, and nothing in it describes what any UAE authority requires or permits in a commercial agreement. Nothing here is a statement of what any provider charges, ours included, and nothing here is a market rate, an average, a benchmark or a guide to one. Nothing here makes a business shown, named, cited or recommended in any AI product, and nobody can promise that it will.

Who publishes this. AI Visibility is owned by Lunasol. The owner is a Dubai agency that sells the kind of work this article describes, and its own GEO and AEO scope is published on the page cited in the example above. Because the owner sells this work, the article carries no price at all. The example scope is quoted only from what that page already publishes, with nothing added and no figure of any kind. We name no supplier from the live check, because naming companies we did not examine would turn an illustration into a review of them. We name our own owner because a reader is entitled to know who publishes this.

The live check was one tool, one day, one connection, two questions and one run each, in Google's AI Mode. Link counts were read from the rendered page rather than by clicking. It is an illustration of what a buyer is told when they ask, not a survey of the market, not a measurement of what anybody charges, and nothing in it is a finding about any supplier. We name no agency, publication, product, host or domain from either answer. The two figure ranges we found are described by their ratio and their attribution and are never reproduced, because a figure with no authority behind it becomes an anchor the moment a reader sees it. We did not attempt a second assistant: this environment produced no usable answer from it across three earlier articles in this series, and that decision was taken before these runs rather than after them. The logged runs are available on request, in a form carrying no company, product or domain name. Assistant answers vary by account, location and time, and quoted 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 is a product of OpenAI, and the OpenAI Help Center is a publication of OpenAI. Google Search, Google Search Central, Google Search Console, Google Analytics, AI Mode and AI Overviews are products or publications of Google. WhatsApp is a product of Meta. All are trademarks of their respective owners.

AI Visibility is owned by Lunasol.