Website GEO for ChatGPT and AI Overviews: access, name and tracking
Website GEO is work to make sure ChatGPT, Perplexity, Gemini, Copilot and Google's AI Overviews can read your pages, do not mix your company up with a namesake, and have something to quote. We open access to agents that come on a person's request, separate your name from other entities, rewrite the first paragraphs of key pages, and set up tracking of referrals from assistants. We do not promise a mention in the answer: no vendor guarantees one.
referrals in GA4, AI agents in server logs, Google and Bing reports on AI impressions
after a free audit
Price
3012345678900123456789
Guarantee
days after project sign-off
referrals in GA4, AI agents in server logs, Google and Bing reports on AI impressions
What we measure
1012345678920123456789–2012345678970123456789
Timeline
working days, then a follow-up snapshot after 30 days
a free AI visibility snapshot, 1–3 working days
Before the estimate
6012345678920123456789
Measured example
sessions from chatgpt.com for 1 July – 2 August 2026 on our service kvednace.com.ua
a mention in an assistant's answer: the vendor decides which sources to pick
What we do not promise
Who it's for
Situations where this service delivers results
Scenario 1 of 4
You are chosen after reading, not on the first click
Services, B2B, medicine, expensive equipment: a person first asks an assistant how to choose and what to look out for, and only then looks for a provider. If the site does not answer those questions in plain text, the assistant will take the answer from other sites.
We'll review your situation in a free audit
Scenario 2 of 4
Someone else has your name
We have been through this ourselves: the short name “LIONEX” is shared by at least three unrelated companies, and when asked about it, a model answered about a crypto exchange. Code alone cannot fully resolve such an overlap, but the names can be told apart in markup and text, and a list of external steps can be drawn up.
Scenario 3 of 4
A store with a catalog of thousands of items
For an assistant, a product card is a paragraph from which an answer either gets extracted or does not. Descriptions go through Textory's GEO mode, where each text gets a 0–100 score for structure, and you can see which categories lag behind.
Scenario 4 of 4
chatgpt.com has already appeared in your GA4
The source is there, but it is unclear what to do with it: which pages people open, where they come from, whether it is worth investing in. We start with tracking so the decision rests on a number.
Cases
Tasks and results in numbers — all metrics measured by us
our free public service for entrepreneurs: a converter from KVED-2010 codes to NACE 2.1-UA
Task
Make it possible for an entrepreneur to find their new code straight from search, without going through the service's home page.
Solution
A separate page for each code, open indexing with separate rules for AI assistant bots in robots.txt; session sources are visible in Google Analytics.
Result
For 1 July – 2 August 2026 GA4 recorded 62 sessions from chatgpt.com — the third source after Google search (217) and direct visits (196), ahead of Bing (30). The sitemap has 1,366 URLs, 616 of them codes on their own pages; the server's first response was 244 ms, measured 2 August 2026. There was no baseline before launch, so the contribution of the robots.txt rules to these 62 sessions has not been measured. The limit: these are referrals to one service over one period, not a share of mentions and not a forecast for another niche.
a plastic surgery and aesthetic medicine clinic
Task
A clinic website with a blog and online booking; separately — what search and AI agents receive, and which AI agents to let in.
Solution
Prerendering for bots, clinic and doctor markup, llms.txt as a content map, robots.txt with a split policy for AI agents.
Result
robots.txt blocks seven training agents, including GPTBot, ClaudeBot and Google-Extended, and admits six search agents that cite their sources, including ChatGPT-User, OAI-SearchBot and PerplexityBot. Measured 31 July 2026. The limit: agent visits and referrals from assistants were not measured on this site — the groundwork is there, the return is unknown.
What's included
Complete list of work and what you get as a result
We check which AI agents can reach the site: robots.txt, bot protection, CDN
We separate agents that collect training data from those that work for search and answers: you decide whom to let in. For ChatGPT-User, OpenAI warns that robots.txt may not apply: the agent acts on a person's request
We set up a separate GA4 channel group for assistants — referrals from them no longer get lost among other referrals
We connect the AI Performance report in Bing Webmaster Tools and the AI impressions report in Search Console — figures from the vendors themselves
We count AI agent visits in the logs, separately ChatGPT-User, Claude-User and Perplexity-User, which come at a person's request
We take a baseline: 15–30 of your clients' questions in four assistants, with the date and the verbatim answer — a sample, not a ranking
We separate the company from its namesakes: Organization with a single @id, alternateName, sameAs, the founder as a separate entity, a “who we are” paragraph as visible text
We rewrite the first paragraphs of key pages: the answer in the first sentence, followed by figures and terms that can be verified
For a catalog — Textory's GEO mode: each description gets a 0–100 score, and you can see which categories lag behind
We write llms.txt — a cheap step we do not pass off as a lever
We fact-check every claim of a GEO audit before changing anything
We hand over a list of external steps the website cannot cover: company profile, directories, Wikidata
When this service isn't right
What's not included — so there are no surprises at delivery
A promise of a mention in ChatGPT, Perplexity or AI Overviews
Inflating mentions or paid “contractor rankings” for the sake of citations
Mass generation of pages for assistant questions
Moving the site to server-side rendering — that is separate work
Running your external profiles: we provide the texts, the accounts stay yours
Process steps
Transparent stages with approval at every step
Total duration:11–25 days
01
Snapshot and baseline
3–5 working days
Agent access, the server's first response without JavaScript, organization markup. In parallel — 15–30 of your clients' questions in ChatGPT, Perplexity, Gemini and Copilot, with the date and the verbatim answer.
02
Channel tracking
2–4 working days
A GA4 channel group for assistants, counting AI agents in the logs with a separate line for those that come on a person's request, the AI Performance report in Bing Webmaster Tools and the AI impressions report in Search Console. Without this step the rest of the work has nothing to be compared against.
03
Access and map
1–2 working days
Agent rules in robots.txt as you decide, a check on the live server, llms.txt. If it turns out that the content is assembled only in the browser, we stop and suggest rendering first.
04
Entity
2–5 working days
Organization with a single @id, alternateName, sameAs, the founder, a defining paragraph on the “about us” page. Plus a list of external profiles to create or fix, with ready-made texts.
05
Pages that can be quoted
3–9 working days
The first paragraphs of key pages in an “answer first, then facts” format, agreed with your specialist. For a catalog — a pass of descriptions through Textory's GEO mode with a score for each.
06
Follow-up snapshot
1–2 working days
30 days after the previous stages — the same questions in the same assistants, referrals in GA4, agent lines in the logs, and the Google and Bing reports. The report shows what has changed and names plainly what cannot be seen after a month.
Free AI visibility snapshot
We look at the site the way assistant agents see it and check whether you have any tracking of this channel at all. No access is needed for the first snapshot — only the site address and 5–10 questions your clients ask you.
What we measure
Agent accessWhich AI agents robots.txt lets in, and whether bot protection on the hosting cuts them off.
What an agent gets without JavaScriptThe server's first response on key pages: does it contain text, or is it an empty shell?
How you are recognizedOrganization markup, the full name, overlaps with namesake companies in assistants' answers.
Whether there is anything to quoteThe first paragraphs of key pages: do they answer a question or greet the visitor.
Channel trackingWhether analytics shows referrals from assistants separately or they sit among referrals and direct visits, and whether the AI impression reports in Search Console and Bing Webmaster Tools are connected.
What you get
A short document: what blocks agents, what confuses your name, which pages have nothing to quote.
Four assistants' answers to your questions, dated — the first baseline.
A conclusion on whether you should work on GEO now or on regular SEO first.
A 30-minute call to go through the document.
Timeline: 1–3 working days
Why is it free
The snapshot is done from outside and fits into 1–3 working days. Some sites hear afterwards that it is too early for GEO — it is more honest to say that before the estimate than after.
What's next
Then an estimate by stage: tracking, access, entity, pages. You can take tracking alone and decide on the rest once the first numbers appear.
Short form: your contact and site URL
We work officially
Contract, act and invoice — for every project, not just large ones.
Every project gets a written contract: scope, deadlines, amount, acceptance procedure. After delivery — act and invoice, then 30 calendar days of warranty.
02
Sole proprietor & bank transfer
The contractor is a registered sole proprietor. Payment by invoice with closing documents.
03
Rights & access — yours
Code, design and materials transfer to you after full payment. Domain, hosting, repository and analytics are registered to you.
04
Client portal instead of email chains
During the project you get access to a portal: contracts, invoices, acts and project status in one place.
05
Sites in nine countries
Our sites run in nine countries: besides Ukraine — Poland, Germany, Austria, England, Norway, Bulgaria, Moldova and Spain.
06
Verifiable numbers
Every case in the portfolio comes with a link to a live site and a technical measurement.
07
Audit first, then pricing
There is no price list on the site intentionally: the scope of the same work differs multiples between clients.
08
We say "no" when unsure
If the task isn't ours or the deadline is unrealistic — we tell you upfront.
01
Contract, act and 30-day warranty
Every project gets a written contract: scope, deadlines, amount, acceptance procedure. After delivery — act and invoice, then 30 calendar days of warranty.
02
Sole proprietor & bank transfer
The contractor is a registered sole proprietor. Payment by invoice with closing documents.
03
Rights & access — yours
Code, design and materials transfer to you after full payment. Domain, hosting, repository and analytics are registered to you.
04
Client portal instead of email chains
During the project you get access to a portal: contracts, invoices, acts and project status in one place.
05
Sites in nine countries
Our sites run in nine countries: besides Ukraine — Poland, Germany, Austria, England, Norway, Bulgaria, Moldova and Spain.
06
Verifiable numbers
Every case in the portfolio comes with a link to a live site and a technical measurement.
07
Audit first, then pricing
There is no price list on the site intentionally: the scope of the same work differs multiples between clients.
08
We say "no" when unsure
If the task isn't ours or the deadline is unrealistic — we tell you upfront.
We sign NDAs on request. After completing the work, we delete our access to your services ourselves.
Did not find your case?
Describe how it works on your side — we will tell you whether “Website GEO optimization” fits and what it means in your situation. No brief and no call: one question, one answer.
What affects the price
Why two seemingly identical tasks are priced differently
Number of key pagesThe first paragraphs of five service pages are one job, fifty categories are another. We take the ones clients come through.
How crowded your name isA unique name is separated quickly through markup. If several companies share it, a defining paragraph, profiles and a list of external steps are added, and the timeline grows.
Access to server logsWhen logs exist, AI agent tracking is set up on the data already there. When the hosting does not provide them or keeps them only for a day, we first switch on collection and wait.
Catalog sizeProduct descriptions go through Textory, and the processing cost is set by the volume of text. A thousand cards and fifty thousand are different estimates even with the same logic.
Language versionsEach language means separate first paragraphs and a separate snapshot of answers: in Ukrainian and in English an assistant may quote different sources.
Technologies & integrations
What we build on and what it connects to
Stack
GA4, a channel group for assistants — shows referrals, not mentions; a referral from an app may arrive without a source
Server logs with the user-agent string — this is where ChatGPT-User and Perplexity-User show up
robots.txt with split rules: the vendors themselves document the agent names
JSON-LD Organization and Person with @id, alternateName and sameAs
llms.txt — an optional format: Google states plainly that Search does not use it, and we have not seen commitments from other vendors to read it
Textory, GEO mode — a 0–100 description score by fixed rules, with no model in the scoring itself
Search Console — the AI Overviews and AI Mode report, for all sites since 31 August 2026: impressions by page and country, with other metrics Google promises to add later; Bing Webmaster Tools, AI Performance (public preview) — Copilot citations by page and the queries for which Copilot looked for sources, without clicks
Integrations
Google Analytics 4
Google Search Console
Bing Webmaster Tools
Textory
server logs
Tracking and analysis versus a package promising mentions
How this option differs from the alternative
The “you will get into ChatGPT answers” packageOur approach
What is promiseda mention in an assistant, which the contractor does not controla defined scope of work and channel tracking
How the result is checkeda screenshot of one lucky answerGoogle and Bing reports, referrals in GA4, agents in logs, a dated snapshot of the same questions
A name others shareusually goes unaddressedwe separate it in markup and text, and give the remaining steps as a list
Where the texts come fromdozens of new pages for assistant questionsthe first paragraphs of existing pages, based on your facts
When they say “too early”after payment, if at allin the free snapshot, before the estimate
What we need from you
We can't start without this — best to prepare in advance
1The site address and 5–10 questions clients ask you before buying — the snapshot is built from them.
2Access to GA4 with edit rights: the channel group is set up in the property itself.
3Access to Search Console and Bing Webmaster Tools, or consent to verify the site in Bing if it is not there yet.
4Access to server or hosting logs. If there are no logs, we switch on collection at the start.
5A list of the company's profiles outside the site — directories, social networks, registries — and who has access to them.
6A specialist who confirms the facts in the first paragraphs: figures, terms, timelines.
If something is missing — let us know, we'll help you gather it or do it as a separate task.
FAQ
Most frequently asked questions — with concrete answers
What is GEO and how does it differ from SEO?
GEO — generative engine optimization — is work to make sure assistants can read the site, recognize the company correctly and quote a specific paragraph. The foundation is the same as in SEO: AI Overviews links only to pages in Google's index that are eligible to show with a snippet; Copilot finds sources through Bing search. What is new is agent access, separating your name from other entities, and tracking the channel: referrals in GA4, agents in logs, and the AI impression reports Google and Bing opened up in 2026.
Do you guarantee that ChatGPT will mention us?
No. The vendor chooses the sources for a given answer; Google publishes the eligibility condition (the page must be indexed and eligible to show with a snippet) and general advice, but not the rules by which it picks a source for an answer. The same query from two people can produce different answers. We are responsible for what can be checked: agents have access, the name is not confused, key pages have a paragraph fit for quoting, and the channel is visible in analytics.
How do we know it is working?
Through four signals, and none of them gives a share of mentions across all assistants. The first is vendor reports: Bing Webmaster Tools counts Copilot citations by page, and Search Console has shown impressions in AI Overviews and AI Mode since 31 August 2026 — impressions only, for now. The second is referrals from assistants in GA4: on our service kvednace.com.ua for 1 July – 2 August 2026 that was 62 sessions from chatgpt.com, the third source after Google and direct visits. The third is visits by agents acting on a person's request, in the server logs. The fourth is a dated snapshot of the same questions. On our own site, the first 10 hours of tracking on 26 August 2026 showed zero language-model agents out of 254 bot requests, and we show figures like that too.
Should we block GPTBot or not?
That is your decision, and it is more than one decision. OpenAI itself separates three agents: GPTBot collects training data, OAI-SearchBot is needed for the site to appear in ChatGPT search, and ChatGPT-User opens a page when a person asks it to. You can block the training one and let the rest in — that is how it is set up on the site of a clinic from our portfolio. For ChatGPT-User, OpenAI warns that robots.txt may not apply: the agent acts on a person's request. One more limit: blocking the training agent does not remove your text from models that have already been trained.
Will llms.txt help?
As a cheap step, yes; as a lever, no. It is a concise map of sections for agents, but the format is optional: Google states plainly that Search does not use it, and we have not seen commitments from other vendors to read it. We put it last, when everything else is done.
Our name gets confused with another company. What can be done?
We have tested this on ourselves: the short name “LIONEX” is shared by at least three unrelated companies, and in August 2026 a model asked about it answered about a crypto exchange. So wherever confusion matters, we write the full name — LIONEX AI First Agency, which nobody shares. The website closes this only partly: one organization with an @id, the full name and alternateName, sameAs pointing to the real profiles, a defining paragraph as visible text. The rest is external signals such as a company profile and directory listings, and that takes months.
When will results be visible?
The structural work takes 12–27 working days, followed by a follow-up snapshot after 30 days. When assistants will start picking up your pages, and whether they will at all, cannot be predicted: each vendor refreshes its sources at its own pace. So the report after a month names plainly what has changed and what is not visible yet.
What does Textory do for an online store?
Its GEO mode rewrites descriptions from your specifications so that the first sentence answers the question and the facts sit under question-style subheadings. Each text gets a 0–100 score by fixed rules — direct answer, fact density, entities, list of specifications, length; no model takes part in the scoring itself. The values 77 in the preview and 82 on the product card are examples from a demo catalog, 5 September 2026. The score measures structure, not the probability of a mention; the mechanics are covered in detail in the blog article “GEO optimization of descriptions: how a product gets into AI search”.
Services show a “share of voice in AI”. Isn't that the same thing?
Similar, but different. Such services run their own set of queries and count how often your name comes up in the answers. As a trend it is useful, but the service picks the queries, and they may not match your clients' questions. Bing and Google reports count your real impressions and citations, but each only within its own AI features. If you already have a share-of-voice service, we will use its data in the report, but we will not derive market share from it.
This is how your store can look on a phone: a collection banner, popular products and a cart one tap away. Tap “Add to cart” and the counter in the header updates, just like in a real store.
Collection bannerThe first screen shows what is new, not a logo.
Product cardsPhoto, price and a button — no extra page to open.
One-tap cartThe counter in the header updates instantly, like in a real store.
iThis is a demo interface, not a client's store. Live projects are in the portfolio.
9:41
SHOP2
New collectionAutumn / WinterExplore
PopularAll products ›
Essentials Hoodie1 290 ₴
Urban Backpack1 490 ₴
Classic Cap690 ₴
Basic Tee590 ₴
Street Sneakers2 390 ₴
Active Shorts890 ₴
Added to cartClassic Cap
Card payment1 290 ₴ · paid
Nova PoshtaWaybill created
Send us your website address and five questions clients ask you before buying.
In return — a snapshot: which AI agents reach you, whether your name gets confused, which pages have nothing to quote, and four assistants' answers to your questions, dated. If it is too early for GEO for you, that will be the first line.
There is no price list on the site on purpose: the same work differs several times over between two clients, and a “from” figure explains nothing in that case. First a free audit — we count your pages, duplicates and speed — then we name the sum and the deadline and fix both in the contract.