GEO packageKreativ Kontroll (Creative Control)

AI search visibility, done for you. Not audited. Fixed.

Your competitor is in the AI answer. You are not.
We fix that for a fixed fee, in a few days.

ChatGPT, Gemini, and Google AI now answer instead of listing. We put the machine readable layer on your site that those systems read before they decide who to name. No developer needed on your side. One payment, no retainer.

Fix my site for 399 EUR →

Card payment. If your GEO Index does not improve by at least 10 points, you get the full fee back.

AI assistant · illustration

> who is the best supplier for this in Europe?

Based on the sources I can read, three companies stand out:

  1. 1competitor-a.comreadable
  2. 2competitor-b.comreadable
  3. 3competitor-c.eureadable
  4. 4your siteno data the model can parse
Sources used for this answer
competitor-a.comcompetitor-b.comcompetitor-c.eu
There is no page two of an AI answer.
3,630 sites measured 43.9 average score out of 100 12 monthly re-measurements included 10 points or your money back
01The situation you are in

Ten blue links used to be a ranking. One AI answer is a verdict.

You know how this used to work. You ranked somewhere on page one, a share of people clicked, and if you slipped to position nine you still got traffic. Position was a dial, not a switch.

That is not what happens now. Someone asks ChatGPT for the best supplier in your category. They get three names and a short paragraph. They do not scroll, they do not compare, they do not open a tenth tab. They call one of the three.

If you are not one of them, you were not ranked lower. You were not there at all.

What the result page used to be, and what it is now
On the left ten ranked links fading down the page, on the right a single answer box that names three companies then: ten ranked links Google ten of these, and position was a dial: slip a place, lose a share now: one answer competitor-a.com competitor-b.com competitor-c.eu no fourth name is offered it is a switch: named, or not named

Schematic drawing of a result page, not a screenshot. Google and the Google logo are trademarks of Google LLC, shown to identify the search interface and nothing else.

Sixty eight percent of Google searches now end without a click. The answer stays in the interface, the visit never happens. That share is not coming back down.
02Why this is not an SEO problem

Your SEO agency is optimizing for a ranking. The AI is not ranking you.

A search engine crawls your page, judges relevance, and puts you in an order. A language model does something else: it needs to state a fact about you, so it looks for facts it can trust in a form it can parse without guessing.

If your company description exists only as a sentence inside a hero image, a human reads it and a machine does not. If your services live in a slider, a human clicks and a machine sees nothing. If your robots file quietly excludes AI crawlers, which happens by accident far more often than on purpose, then nothing else you do matters.

This is why good looking sites score badly, and why the fix is not a redesign. Nine specific things decide whether a machine can describe you correctly. We put all nine in place.

Two different jobs: storing a page against stating a fact
Left column, what a search crawler stores. Right column, what a language model needs before it can name you A SEARCH CRAWLER STORES page text, in any layout links pointing at the page how the page performs an order, relative to others Output: a position on a list. Being ninth still returns something. A LANGUAGE MODEL NEEDS what you do, declared as data who you are, as one entity permission to read the site your wording, in quotable form Output: a sentence with your name in it. Missing data means no sentence at all.
62out of 100

basecamp.com

A software company with a good site, a clear message, and thirty years of writing about how to do this well, measured on the same meter we point at your site. It scores 62. The gap is not competence. Almost nobody has done this layer yet, which is exactly how early this market still is.

03How the work is done

Nine checks, one fixed scope, a few days

We measure your homepage against the nine items an AI search system uses to build an answer about you, then we implement the ones you are missing. Same nine every time, which is exactly why the price can be fixed.

Structured data

Schema and JSON-LD describing your organization, your location, your site, and your navigation, in the format these systems parse first.

A machine states what you do instead of inferring it from your page layout.

llms.txt

A dedicated file telling AI crawlers which parts of your site matter and how to read them.

You decide what gets picked up, rather than leaving it to chance.

Crawler access

Robots rules that explicitly allow the AI crawlers you want, and reveal the ones you were blocking without knowing.

The systems you want to appear in can actually read the page.

Question and answer structure

An FAQ layer built the way a model reads questions, so your own wording is what gets quoted back.

Your answer competes for the citation, instead of your competitor's blog post.

Identity signals

Meta layer, heading hierarchy, and verified sameAs links to your real profiles.

The model connects the same company across sources, instead of treating you as three unrelated mentions.

Your score, before and after

You get the number, not an opinion. Run it again yourself any time with the Chrome extension included in the package.

You can verify our work without taking our word for it.

Three competitors, measured by name

You name three competitors. We run the same nine checks on their sites and put the scores in your report, next to yours, with the domains named.

You find out whether you are behind the market or ahead of it, which your own score alone cannot tell you.
The nine items, all nine implemented, same list every timeopen any item to see what it is

The nine are not our invention and they are not a menu. They are the nine places an AI search system looks when it has to state a fact about a company: four of them declare what you are (structured data, the organization and location entities, your identity links), three of them decide whether a machine may read you at all (crawler permissions, the llms.txt file, the meta and heading layer), and two of them turn your own wording into something quotable (the FAQ structure and the answer formatting). Miss one and the model guesses. Miss three and it names somebody else.

Every project gets the same nine, measured the same way, which is why the fee is a single number rather than a range.

9 / 9 implemented, not listed. The scope does not grow with the site, which is why the fee can be a single number.
1Schema, JSON-LD

A block of JSON-LD in your page source that states, in a format a machine parses without guessing, what this page is and what this company is. Your visitors never see it. It is the first thing an AI system looks for, and when it is missing the model has to infer your business from your layout, which is exactly where wrong answers come from.

2Meta tags

The title, description and canonical tags that say what the page is and which address is the real one. They also decide what your page looks like when something quotes or shares it. Duplicated or missing ones make one site look like several unrelated pages to a parser, so the facts never add up to a company.

3Heading structure

One H1 that states what the page is, with H2 and H3 in order underneath it. Machines read the heading tree as the outline of your content, so an outline built out of styled divs arrives as one undifferentiated block. This is the cheapest item on the list to put right and one of the most frequently broken.

4FAQ section

Real question and answer pairs on the page, marked up as FAQPage structured data. This is the format assistants quote from most readily, because a question with one clear answer is already the shape of a reply. It is also the one place where you choose the exact wording that gets repeated about you.

5llms.txt

A plain text file at the root of your domain, next to robots.txt, written for language models rather than people. It summarises who you are, what you sell, and where the pages that matter live. It is the only item here that exists purely for machines, it costs nothing to serve, and it removes the guesswork from the first read.

6AI crawlers, robots

The rules in robots.txt that decide whether GPTBot, ClaudeBot, PerplexityBot, Google-Extended and the rest may read you at all. Blocking them is almost always an accident, inherited from a template or a security plugin. If this one is wrong the other eight cannot rescue you, because nothing ever gets read.

7Organization, LocalBusiness

The entity types in your structured data that say this is a company, this is its legal name, this is its address, these are its contact points. Search systems store entities, not pages, and this is what your facts get attached to. Without it a model holds a document about you rather than a record of you.

8sameAs profiles

A list inside your structured data of the profiles that are genuinely yours: LinkedIn, the company register entry, Facebook, YouTube, the industry directory. It lets a machine connect this site to the entity it already knows from elsewhere. That link is what turns scattered mentions into one confident answer.

9AI visibility

The outcome check: we ask real assistants the questions your buyers would ask, then record whether your domain is named in the answer. It is the only one of the nine that measures the result rather than the plumbing. It is also the number that tells you whether the other eight did their job.

WordPress
WordPress: we install our plugin. Shopify, Shoprenter, UNAS, or a custom stack: we implement it manually or by script.

The package works on any platform, only the installation method differs. Three competitors, not four: it is a fixed scope, like everything else here.

04What you are probably thinking

The four questions we get before anyone pays

I already pay an SEO agency. Is this not their job?

It is adjacent to their job, and in practice they are not doing it. Ask them for your llms.txt file and your Organization schema. If the answer is a proposal instead of a URL, it is not in place. This work is a one time technical layer, not a monthly retainer, so it does not compete with what they do and it does not replace them.

What stops me from having a developer do this?

Nothing, and if you have a developer with time, that is a legitimate choice. What you are buying here is the specification and the certainty: the nine items, in the formats these systems actually accept, plus the measurement that tells you it worked. Most teams do not fail at the code. They fail at knowing which nine things, in what form, and then at verifying it afterward.

Every provider I contacted quoted a range. Why is your price a single number?

Because the scope is a single number too. Nine checks, one domain, implementation included. We are not estimating unknown hours, which is what a range really admits. If your site needs work beyond those nine items, we say so before you pay, and that work is not part of this package.

How do I know the score will not just drop back next month?

Some of it can, and we will not pretend otherwise. AI crawler standards change, and a site edit can remove an element we added. What we install is stable, but it is not immune to your next redesign.

This is exactly why we re-measure your site every month for twelve months and email you the result. If an element disappears after a redesign, that report is where you find out, and you find out from us rather than from a drop in inquiries. It is included in the fee. Continuous prompt-by-prompt tracking is a different service, and most companies buying this page do not need it.

05What you get to keep

Everything we install stays yours

There is no license to renew and nothing switches off if you leave.

What sits on your domain after we leave, in your own files
A file tree on your own domain holding the structured data, the llms.txt file, the robots rules, and the FAQ layer, with nothing left on our side YOUR DOMAIN yourdomain.com/ llms.txt robots.txt <script type="application/ld+json"> faq markup, in your pages OUR SIDE nothing to renew nothing to switch off
06What changes, realistically

What actually looks different in a month

We are not going to promise you revenue. Here is what the mechanism can and cannot do.

What the layer does

  • An AI system asked about your category can state what you do, in your words, because the description now exists in a form it can read.
  • Your own FAQ wording becomes quotable, so when someone asks a question you already answer, your answer is a candidate.
  • The crawlers you were accidentally excluding can read the site, which is often the single largest jump in the score.
  • Your score is a number you can re-run yourself, so the question is settled by measurement rather than by opinion.

What it does not do

  • It does not write content for you.
  • It does not build links.
  • It does not put you in an answer where you have nothing relevant to say.

It removes the technical reason for being left out. The rest is your business.

07What you are actually buying

You are not buying a report. That is the entire point.

Every provider we measured in this market sells the same thing first: an audit. You pay for a document that tells you what is wrong, and then the repair is a second conversation with a second, larger, open ended number attached.

We inverted that. The measurement is free and takes half a minute. The thing you pay for is the work being finished. When the invoice arrives, the files are already on your site.

The same job, sold two ways
The usual route ends with a second, open ended quote before any work happens. Here the measurement is free and the payment comes after the work is finished THE USUAL ROUTE you pay for an audit you get a report a second quote, open ended work, maybe, later two decisions, two invoices, and the number you care about is the one you have not seen yet THIS PAGE the measurement is free all nine items implemented then the invoice, one number one decision, one invoice, and the files are on your site before it arrives
08The price

399 EUR, once, with the work done

GEO package

implementation, not advice

€399one time payment, per domain
  • The measurement, on the same nine checks every time
  • All nine items implemented on your site by us
  • Installation on your platform, whatever it runs on
  • Your before and after score, in writing
  • The Chrome extension, so you can re-run it yourself
  • A named benchmark against three competitors of your choosing
  • Twelve months of monthly re-measurement, by email

No retainer. No recurring charge. Nothing to cancel.

Fix my site for 399 EUR →

Card payment, EU invoice. Work starts the same week.

Every provider we measured sells you one report. We measure your site every month for twelve months and email you what changed. That is in this fee, not on top of it.

Twelve monthly re-measurements, inside the same fee
One payment at month zero, then a measurement email every month for twelve months pay once one measurement email per month, for twelve months

For context, from public pricing

A one time GEO audit in English speaking markets runs from 500 to 2,000 USD, and monthly monitoring from 500 to 1,500 USD. A Swedish provider lists 4,900 SEK, roughly 430 EUR. In every one of those cases you are buying the report. The implementation is quoted separately, after.

What is not included, stated plainly

Content writing is not included. Link building is not included. Continuous prompt-by-prompt tracking of your AI visibility is a separate service, not part of this package: if that is what you need, tell us and we will explain what it involves. You do not need it to get the fix, and the twelve months of monthly re-measurement described above are included either way.

before after 10points or your money back
09The guarantee

If the number does not move, you do not pay

We measure your GEO Index before we touch anything, and again after the work ships. If it has not improved by at least 10 points, you get the full fee back. Not a credit, not a partial refund, not a discount on the next thing.

This is a safe promise for us to make for an unglamorous reason: on the nineteen sites we have published, the smallest improvement was 35 points. We are guaranteeing well below what the work normally does. That is the correct way around.

10Who this is for, and who it is not for

Let us save you a call

This is for you if

  • You have a working site and you sell services or expertise, in a category where people start with a question.
  • You have noticed competitors being named in AI answers where you are not.
  • You want a defined project with a defined end, not another monthly line item.
  • You need something you can show internally as done, with a number attached.

This is not for you if

  • You do not have a site yet, or you are rebuilding from scratch. Do that first.
  • You are a technical team with time to implement schema, llms.txt, and an FAQ layer yourselves. You do not need us, you need the checklist.
  • You want AI visibility tracked continuously across many prompts. That is monitoring, which is a separate service, not this package.
  • Your site is a large consumer platform behind aggressive bot protection. See the note at the bottom of this page.
11The numbers we have

What 3,630 measurements look like

We have been measuring AI visibility on real sites since 2023, when we first presented on this from a conference stage. The dataset is ours and it is not flattering to the market.

3,630sites measured
43.9average score out of 100
40median score
36pages of our own published research, 2025
Zoltan Sipos, founder of Kreativ Kontroll
Zoltan SiposFounder, Kreativ Kontroll

We first presented on AI visibility from a conference stage in 2023, and we have been running the same meter on real sites ever since. The scoring matrix behind it comes out of our own 36 page research, so a score means the same thing in March and in November.

Where the market sits: score distribution across the measured sites
A distribution curve of measured scores, bunched toward the low end, with the average marked at 43.9 and the fixed range marked near 100 02040 6080100 43.9 average most of the market sits left of this line after the fix our published set GEO Index, 0 to 100. Curve shape from our own measured set.

Read the average again: most sites are failing this, including well funded ones. We measured stripe.com at 78 and basecamp.com at 62. Those are not sloppy companies. The gap is not competence, it is that almost nobody has done this layer yet. That is the window, and it closes as soon as it becomes standard.

12Where this goes after the fix

The fix is a floor, not a subscription trap

Most projects end here, and that is a legitimate outcome. If you want more later, these are the honest next steps, in order of who actually needs them.

Where most projects stop, and what the steps after it actually are
Four ascending steps, with the first one marked as where most clients stop the fix, installed re-measure yourself watch the three prompt tracking most clients stop here only if you have a reason
  1. Nothing, and you still get measured. The layer is installed and it does not expire, and for the next twelve months the monthly re-measurement arrives on its own. Most clients stop here, and nothing about that is a compromise.
  2. Re-measure whenever you want. The Chrome extension is in the package. Use it after a redesign, or whenever your developer touches the theme, without waiting for the monthly report.
  3. Watch the three competitors. You get them benchmarked once, by name, in your report. Re-run those three yourself later with the extension and you can see whether the market moved or you did.
  4. Continuous prompt-by-prompt tracking. If AI visibility is a metric you report internally, tracking it across many prompts over time is a separate service. Ask us when you have a reason to.
13Evidence

Nineteen sites, published with both numbers

Before and after scores from real sites, our own included. The four with the largest movement are in the table, the whole set is summarised under it. Every domain we name can be run through the meter by you, on your own.

What we promise against what the published set actually moved
The guarantee is 10 points. The smallest improvement in the published set is 35 points and the largest is 66 points gained what we guarantee +10 smallest in the set +35 largest in the set +66
domainbeforeaftermovement
230volt.hu34→100+66
wellworking.hu34→100+66
shop.butoralkatreszbolt.hu43→100+57
kreativkontroll.hu our own site30→87+57

Four rows out of the nineteen sites we have published so far. Across all nineteen the average movement is +50.8 points, the smallest is +35 and the largest is +66, against a guarantee of 10. Eleven of the nineteen finished at 100 and eight did not, including two that stopped at 67, which we keep in the set on purpose: a reference list with no ceiling cases in it is a marketing asset, not evidence.

14FAQ

Practical questions

What the first fortnight looks like, and what arrives after it
Access on day zero, the nine items implemented within a few days, the before and after score in writing, then a measurement email every month for twelve months day 0 you hand overaccess a few days all nine itemsimplemented on delivery before and after score,in writing next twelve months one measurementemail a month No discovery phase, no workshop, no weekly status call. You are asked for one thing: access to publish files on your domain.
How long does it take?

A few days from the point we have access. Not weeks, and there is no discovery phase to sit through.

What access do you need?

Enough to publish files on your domain. On WordPress that is an admin account. On other platforms it depends on the stack, and we will tell you exactly what before you pay. We do not need access to anything beyond the site.

Will this slow my site down?

No. What we add is text based metadata and a small file. It does not load scripts on your front end and it does not change how your pages look.

Do I need to change my design?

No. Nothing we install is visible to your visitors. Your layout, copy, and brand are untouched.

What if I move platforms next year?

The structured data and the FAQ layer are content, so they migrate with your site if your new platform supports them. The llms.txt file and robots rules take minutes to recreate, and your implementation notes tell your developer how.

Is one payment really one payment?

Yes. There is no recurring charge on this package, nothing auto renews, and there is nothing to cancel. Monitoring, if you ever want it, is a separate decision you make later.

Do you work in my language?

The implementation is language independent: structured data and robots rules are not written in English or Hungarian, they are written in a format. We handle the project in English and we mark up your existing content in whatever language it is published in.

15Closing call

Measure it first. It takes half a minute and costs nothing.

Put your domain in the field and you will get your score out of 100, plus the specific items you are missing. No call, no discovery session, no sales sequence you have to escape. If the score is already high, we will tell you that and you should not buy this.

What comes back from the free measurement, in about half a minute
A score out of 100 on the left, and the list of the nine checks with the missing ones marked, on the right 44 out of 100, your GEO Index Meta tags Heading structure llms.txt missing AI crawlers blocked Schema, JSON-LD FAQ section missing sameAs missing Organization Illustration of the report layout. Your own result depends on your site.
Free measurement

Where does your site stand right now?

Put in your domain and we run the same meter we run on paying clients. One score from 0 to 100, plus the items you are missing, in about half a minute.

1Your domain. That is all we need to start.

No email at this step. No account, no call, no sales sequence.

Reading your site the way an AI assistant reads it.

    2Your result is ready. Where should we send the copy?

    The score, the missing items, and what each one is worth in points. We email you the same result you are about to see, so you can forward it internally.

    0/100
    +0 points

    available on your site from the items above. We implement all of them for €399, and guarantee at least 10 points or your money back.

    See what €399 covers
    There is nothing here worth selling you.

    Two things are still worth doing. Run the same measurement on the three competitors you would name if someone asked, because your own score alone does not tell you whether you are ahead. Then run yours again in a few months: crawler rules change and a redesign can quietly remove what is in place today.

    Fix my site for 399 EUR →

    Card payment, EU invoice. Work starts the same week.

    16If you are still undecided

    Two things you can do without us

    Run the free measurement on your own site, then run it on the three competitors you would name if someone asked. That comparison answers the only question that matters here: are they in the answer and are you not.

    If you would rather not give us your email yet, do it manually. Open your own robots.txt and search for the AI crawler names. Try to fetch yourdomain.com/llms.txt. Paste your homepage into any public schema validator. Whatever comes back empty is what we install. You will reach the same conclusion we would, it will just take you an afternoon.

    One limitation, stated up front

    Some sites cannot be measured from the outside. Large consumer platforms often run aggressive bot protection that refuses automated requests, and when that happens our meter returns no score rather than a bad one. We have seen it on sites like hemnet.se and bol.com. Our meter reports this honestly as blocked, with the server's response code, so nobody ever receives a fabricated zero.

    What our meter returns when a site refuses automated requests
    Sites that answer return a score, sites behind bot protection return blocked with the response code, never a fabricated zero HTTP 200, MEASURED stripe.com 78 basecamp.com 62 HTTP 403, BLOCKED hemnet.se no score bol.com no score

    For the companies this page is written for, this has not come up. We measured stripe.com at 78 and basecamp.com at 62 without any difficulty. If your site does turn out to be unmeasurable from outside, we will tell you before you pay, because a score we cannot verify is a guarantee we cannot honor.