Mevcut SEO Ajansınız Gerçekten GEO Yapıyor mu?
Somewhere in the monthly report sits this sentence: "AI visibility is up." Underneath it, nothing says which question was asked, which assistant it was put to, or how your brand appeared inside the answer. The sentence may well be true. But as it stands it cannot be verified, and an output nobody can verify is a service nobody can confirm was delivered.
The point here is not to run down your current agency. Quite the opposite: GEO does not have to be a separate budget line at all. It can sit on top of well-built SEO work, be run by the same team, appear on the same invoice. The one thing it does have to be is a separate evidence line. If there is no evidence in the report, the work is not happening — not because the agency is bad, but because that work never entered the scope in the first place.
The honest answer to "can my SEO agency do GEO too?"
The honest answer splits in two: most of the foundation overlaps, while the measurement and question side is a different job altogether.
The overlapping half. Good SEO work already lays GEO's foundation. Pages that carry genuinely readable text, a heading hierarchy that means something, a site that loads fast, an internal link network that builds topical coherence, the brand's name appearing in outside sources, basic schema markup written correctly. None of this was done "for AI", but all of it is needed before a language model can use your page as a source. If your agency does these properly, the first half of GEO is already yours.
The half that does not overlap. Past this point the SEO toolset does not cover the job:
- The unit of measurement changes. In SEO you measure a rank number and a click count. In GEO you measure whether you appear inside an answer. A rank tracker will not show you this, because there is no rank there to track.
- Questions replace keywords. GEO's tracked unit is not a phrase but the sentence a user actually types. Not "teeth whitening price", but "can you recommend a dental clinic in Kadıköy (a district on Istanbul's Asian side) that takes evening appointments and offers instalment payments".
- There is no single engine. ChatGPT, Gemini, Perplexity, Copilot and Google's AI Overviews each behave differently when choosing sources. Appearing in one does not get you into another.
- The output is unstable. Ask the same question twice and the answer can differ. A single screenshot is not evidence; you need repeated measurement and a sample.
- The access layer is a separate job.
robots.txt,llms.txt, server-side rendering, and your CDN's bot policy. These files rarely come up in classic SEO; in GEO they are half the work.
For a fuller account of that split, see our piece on the difference between local SEO and GEO. The short version: your agency is probably building the foundation and not building the measurement or the access layer. That is not an accusation, it is a scope question.
Nine questions to ask your agency
Ask these in a single meeting and request the answers in writing. The aim is not to corner anyone; it is to get down on paper what is in scope and what is not. A good agency will not mind these questions — it will have ready answers to most of them, and for the rest it will say "we don't do that, we can add it to the scope if you want". Discomfort on its own is a signal.
| # | Question | What a good answer looks like | What an evasive answer looks like |
|---|---|---|---|
| 1 | Which prompt set are you tracking, and can I see the list? | A shareable, dated, fixed list of 20-60 questions, broken out by service, city and intent | "We track it but there's no list", or a keyword list dressed up as questions |
| 2 | Which engines are in scope? | Named one by one: ChatGPT, Gemini, Perplexity, Copilot, AI Overviews. What is out of scope is stated just as plainly | "Across all the AIs" — unbounded, and therefore unmeasurable |
| 3 | Does the report show mentions or citations? | The two are counted separately: appearing in the text and being linked as a source sit on different lines | Both are blended into one "visibility" figure, and asking which one is counted does not clear it up |
| 4 | Who writes the schema, which type on which pages? | A table mapping page type to schema type; who wrote it and where it lives is known | "It's switched on in the plugin", with no idea which type appears on which page |
| 5 | Who controls robots.txt and llms.txt, and when did they last change? | Ownership is clear, the last change date and reason are given; if there is no llms.txt, why not is explained | No awareness that the files exist, or "we'd have to check with the host" |
| 6 | Are AI bots blocked at the CDN or security layer — have you checked? | The panel has been opened, the bot policy reported with a screenshot, tests run with bot user-agents | "We didn't know that setting existed", or it has simply never been checked |
| 7 | Who owns the content produced and the schema work? | It is in the contract: copyright with the client, the agency holds a licence to use | A verbal "of course it's yours" with nothing in the contract behind it |
| 8 | On termination, what stays with us? | Access, files and record handover are defined clause by clause; the handover window is set | "That won't happen", or the subject being brushed aside |
| 9 | Which part of this month's work was GEO, in hours or line items? | An itemised breakdown: this many hours on measurement, this many on content, this many on technical | A single line reading "SEO/GEO services" and an amount nobody can break down |
Of the nine, the three most often skipped are 5, 6 and 9. The first two are technical: an agency that never looks at the access layer is talking about visibility without knowing whether your site is even open to AI crawlers — and a single default setting in the security layer can make a month of content work invisible. The ninth is commercial: GEO does not have to be a separate invoice, but it does have to be a separate line. Work that cannot be broken out is work that cannot be measured. For how these questions fit into a wider framework when you are choosing a supplier, we went through it in detail in our guide to picking a GEO agency.
Four pieces of evidence to look for in the report
Even when the answers are good, the real verification happens in the report. Four things to look for.
The prompt list itself. Attached to the report: a fixed, versioned list of questions. It matters that the list does not change month to month; if it does, which question was added when must be written down. A list that keeps shifting makes comparison impossible and leaves the door open to picking the months that look good.
Dated records. Screenshots, or the answer text itself, carrying the date, the engine it was asked on, and the exact wording of the question. An undated screenshot does not count as evidence, because without knowing when it was taken you cannot know what changed.
A before/after comparison on the same questions. Last month you appeared in 6 of 40 questions; this month you appear in 11. The comparison only means anything if the question set is fixed. And "appears" should be counted separately from "linked as a source"; the two are worth different amounts.
An itemised list of the technical changes made. Date, which page, what changed. "Content optimisation was carried out" is not an itemised list. "On 12 August FAQPage schema was added to the service pages and an FAQ block was written on six of them" is. That list also becomes the backbone of any future handover.
If all four are there, the work is being done even if the results have not arrived yet. If none of them are, then however good the numbers look, you do not know what is working.
Red flags
Some of these mean nothing on their own, but when they pile up in the same report it is worth stopping to think.
A GEO success claim backed by a Google ranking screenshot. A rise in classic search results is not proof that you appeared in an assistant's answer. The two measurements may be related, but one does not stand in for the other.
Guarantees. A promise along the lines of "we'll get you to number one in ChatGPT" is a sentence with no technical meaning behind it; there is no fixed rank in generative answers, and the answer can change between two identical questions. Anyone offering a guarantee is committing to something they cannot measure.
Refusing to share the prompt set. That list is not a trade secret; it is the questions your customers are asking. There is no reasonable ground for withholding it.
The "the algorithm is secret" defence. The algorithm being secret does not require the work to be secret. What is being kept from you is not the algorithm, it is the work itself.
Undated images and round numbers. Percentages with no source, a "40% increase" that never says which question set it came from. Numbers are not the problem; unsourced numbers are.
When the agency is good but is not doing GEO
This is more common than you would think: the agency does its job well, the relationship is healthy, and this particular work is simply not in scope. There are three routes, and each of them is the right one at some point.
A scope addendum. If the agency is technically capable, has worked with you a long time and is willing to learn, this is the cheapest route. What you add to the existing contract is a measurement and reporting clause: prompt set, engine coverage, a monthly record archive, and an audit of the access layer. The budget increase usually stays modest, because the foundation work is already being done. We covered the money side of this split in our piece on GEO versus SEO and how to divide the budget.
A second supplier. If your current agency is strong on content and classic SEO but does not want to move into measurement and technical access, handing that second piece to a separate supplier makes sense. There is one condition: who touches what has to be written down at the outset. If two parties are both editing the same robots.txt, trouble is a matter of time. Draw the line at the level of files and panels, and put a joint monthly meeting in the calendar.
Replacing them outright. This is the right answer in two cases only: when the underlying work is also poor (slow pages, weak content, accumulated technical debt), or when there is a transparency problem. With an agency that will not share the prompt set, keeps access under its own accounts, and closes down questions with "the algorithm is secret", the problem is not scope — it is the relationship. Switching has a real cost, so do not rush the decision; on whether to run this work inside or outside, our in-house team versus agency decision matrix puts the options side by side.
Handover and transition checklist
If you have decided to switch agencies — or you simply want to reduce the risk — what needs handing over falls into four headings. This list is far easier to put together while the relationship is good than while it is not.
Access. Domain registration and renewal responsibility, DNS management, the hosting panel, the CMS admin account, Search Console ownership (your account should own it, not the agency's), the analytics property, Bing Webmaster Tools, the CDN and security panel, email authentication records. The rule is simple: the owner of every account should be your company email address, with the agency added as a user. When ownership sits with the agency, the transition turns into a negotiation.
Files. The current robots.txt and llms.txt, how the sitemap is generated, schema templates, the redirect (301) list, source files for the content and the original images. The redirect list matters especially: lose it and the index value of your old URLs goes with it.
Accounts and records. Map listing, industry directory listings, social profiles, the accounts for tracking and reporting tools. Add to those: the latest version of the prompt set, the monthly measurement archive, and the change log. Without those three the new supplier starts from zero and any comparison with the past is gone.
Knowledge. The reasoning behind the work: why these topics were chosen, which experiments did not pay off. This is usually the thing nobody asks for and the most expensive thing to lose; half an hour in the handover meeting beats months of rediscovery.
So what this piece gives you is not a verdict but an audit instrument. Ask the nine questions, ask for the four pieces of evidence, lay the answers side by side. If your agency comes out well, you carry on with confidence; if something is missing, you know which gap to close. If you would like to go through your own report together, get in touch.
Frequently Asked Questions
Can my SEO agency do GEO too, or do I need a separate agency?
In most cases it can, because GEO's foundation — content quality, technical infrastructure and authority work — is already SEO's job. What differs is the measurement and access side: running regular measurements on a fixed question set across several assistants, archiving the results with dates, and auditing robots.txt, llms.txt and the CDN bot policy. If your agency is willing to add that second part to the scope, you do not need to go looking for a separate supplier.
How do I verify an agency's claim that "AI visibility is up"?
Ask for three things: the full list of questions being tracked, which engines the results came from, and dated records. Then compare last month's report with this month's on the same question set; if the set changed, the comparison is void. Also ask that "appears in the text" and "linked as a source" be shown on separate lines in the report, because the two are worth different amounts.
What should I do if the agency will not share the prompt list?
First ask why; a reasonable justification is hard to come by, because that list is not the agency's methodology, it is the questions your own customers are asking. If sharing is still refused, make it a condition at contract renewal. In the meantime you can start building your own list: write down the questions coming into your sales team for a month, so you have an independent comparison set of your own.
What am I most at risk of losing when switching agencies?
The three things most often lost are the redirect list, the measurement archive and the reasoning behind past decisions. Lose the redirect list and the value your old URLs accumulated is broken; lose the measurement archive and the new supplier starts from zero with nothing to compare against; lose the reasoning and paths already tried and abandoned get tried again. Ask for all three explicitly in the handover meeting.
Do I have to open a separate budget line for GEO?
No. GEO can be run inside your existing digital marketing budget, and for most businesses that is the right call. What is required is a separate reporting line: how many hours went to GEO that month, which measurements were run, and which technical changes were applied all have to be separable. Work that cannot be broken out cannot be measured afterwards, and cannot be defended.