Be citable in ChatGPT, Perplexity, Gemini and AI Overviews
Generative engine optimization is the work of making a company a source that ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews use and cite when they write an answer. This channel runs on machine readability and entity clarity rather than backlink authority. Results are probabilistic, and our contracts say so.
SCOPE STRIP
Who owns visibility inside generative engines?
Generative engine visibility is not one team's job: server configuration, structured data and content writing all have to change at once. A marketing agency does not touch robots.txt; a software agency does not write the passages. When those three decisions sit with different owners, the work stalls at step one.
| Decision | Marketing agency | Software agency | Baki Bilisim |
|---|---|---|---|
| Who sets the access policy for AI crawlers such as GPTBot, ClaudeBot and PerplexityBot? | Out of scope | Partly: has file access, does not produce the decision | In scope |
| Who builds the JSON-LD graph and the sameAs network that define the company as one consistent entity? | Out of scope | Out of scope | In scope |
| Who measures brand citation inside generative engines and reports it with a date stamp? | Partly: reports, does not define the method | Out of scope | In scope |
Marks: ✓ accountability sits here · ~ partly, the decision sits elsewhere · — typically out of scope. We are comparing categories of supplier, not naming individual firms.
Why delivery and visibility belong in the same contract is explained decision by decision on a separate page.
What happens when a buyer asks an AI assistant about your company?
A generative engine writes its answer from a handful of sources it selects from its own index and from live web search. If your site cannot be crawled, if the main content sits behind JavaScript, or if what the company does is not defined in machine-readable form, a directory listing or a competitor becomes the source instead of you.
DEFINITION
Generative Engine Optimization (GEO)
GEO is the work of turning a website into an entity that generative AI engines use and cite while composing an answer. SEO targets the ranking, AEO targets the answer box, and GEO targets the attribution inside the generated text.
What this means for your company: part of the supplier research now finishes inside an assistant window rather than on a search results page. If your name never appears in that window, you are not among the companies being considered for the shortlist.
Why is being cited different work from collecting backlinks?
In classic search visibility, authority signals dominate. Inside a generative engine the route to the answer is a chain, and the first broken link stops everything after it. However strong your authority is, if the engine cannot read your page you are absent from that answer.
0
There is no dedicated markup, submission form or paid route defined for appearing in Google AI Overviews. Eligibility follows the same rules as the page's ordinary eligibility for Search results.
Source: Google Search Central — “AI features and your website” · Accessed: 2026-07-29 · developers.google.com ↗ (opens in a new tab)
The practical consequence is that there is no shortcut on sale called “AI visibility”. What can be done is to build all four links of the chain as engineering work. The deliverable list below does exactly that.
What exactly is delivered in a GEO engagement?
Seven items are delivered: an AI crawler access policy, a compact llms.txt, an entity model with its sameAs network, citable content blocks, an identity and NAP consistency audit, a snippet permission policy, and a date-stamped visibility monitoring set. The format and delivery week of each item is written into the contract.
| # | Deliverable | Format | When |
|---|---|---|---|
| 01 | AI crawler access policy | Decision note (two scenarios with their consequences) plus an implemented robots.txt |
Week 1 |
| 02 | Compact llms.txt | /llms.txt at the site root, under 6 KB |
Week 2 |
| 03 | Entity model and sameAs network | Entity map plus a JSON-LD graph per page type | Weeks 2–3 |
| 04 | Identity and NAP consistency audit | Inconsistency list plus a correction plan with an ownership column | Week 3 |
| 05 | Citable content blocks | Page template, writing rule set and worked example passages | Weeks 3–5 |
| 06 | Snippet and content permission policy | max-snippet, nosnippet and data-nosnippet decisions |
Week 4 |
| 07 | Visibility monitoring set | Fixed 25-question set, date-stamped record table and method note | Week 5 |
Should you allow AI crawlers?
This is a strategy decision and your company makes it. Allow them and your content can be used as a source inside generative answers; in exchange it is read and summarised. Block them and the content cannot be used in this channel, but the discoverability goes with it. We put both outcomes in writing and implement your decision.
The technical form of that decision is a single file. Below is this site's own decision: open to all of them. We chose that as a deliberate visibility strategy and documented the reason in the file itself.
# AI crawler policy — the decision is the client's, the implementation is ours.
# This site's decision: open to all (a deliberate visibility strategy).
User-agent: GPTBot
Allow: /
User-agent: ClaudeBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Google-Extended
Allow: /
Sitemap: https://www.bakibilisim.com/sitemap.xmlOne distinction matters here: according to Google's documentation, Google-Extended applies only to Gemini apps and Vertex AI grounding, and does not affect ranking in Google Search or eligibility for AI Overviews. These two decisions have to be taken separately — there is no single “switch off AI” toggle.
Does llms.txt actually help?
llms.txt is a proposed file that summarises, in plain text, what a site is and which pages matter. No major generative engine provider has confirmed in public documentation that it uses the file when selecting sources. We publish it, keep it small, and never present it on its own as a visibility promise.
2024
The year the llms.txt proposal was first published. The proposal sets out to give language models a plain-text summary of the site together with its priority links.
Source: llmstxt.org — Jeremy Howard (Answer.AI) · Accessed: 2026-07-29 · llmstxt.org ↗ (opens in a new tab)
Enormous llms.txt files that dump the whole site into one document do not work: the file goes unmaintained, it starts to contradict the real content of the site, and the moment it contradicts anything it stops being a trustworthy signal. This site's own llms.txt file is under 6 KB and contains only the corporate summary, the service and guide links, and the contact details.
Which steps does a GEO engagement follow?
Five phases: baseline measurement, access and machine readability, entity clarity, citable content, and re-measurement. Starting and ending with measurement is deliberate, because work whose starting point was never recorded cannot be proven afterwards. Each phase has its deliverable, its measurement criterion and its accountable party written down in advance.
-
Visibility baseline measurement
A fixed set of 25 questions is derived from sales records, support tickets, search queries and the questions buyers ask during supplier evaluation. The set is run across five engines, and whether the brand is mentioned, which sources are shown and whether the information is accurate are all recorded.
-
Access and machine readability
The AI crawler policy is decided and applied in robots.txt. The main content is verified to be present in the first HTML response without JavaScript, snippet permissions are reviewed, and a compact llms.txt is published at the site root.
-
Entity clarity
A JSON-LD graph of Organization, LocalBusiness, Person and Service nodes is built and the
@idarchitecture is made consistent. The sameAs network and the NAP data are written identically on the website, in the business profile and in directory records. -
Citable content
Question-form headings, 40 to 60 word passages that answer in the first sentence, definition boxes, sourced figures and visible update dates are added. There is a single test: the passage has to stay accurate and complete when it is lifted out of its context.
-
Re-measurement and reporting
The same question set is run again with the same method on the same engines. A date-stamped report comparing the result against the baseline is delivered, and a source-correction plan is produced for answers that state incorrect facts.
Within our six-phase working model, these five phases map onto the production and optimisation steps.
When is the work considered finished?
The work is finished when all seven criteria below have been measured and recorded. Each criterion enters the contract together with its threshold and its measurement tool. If a criterion cannot be measured it is removed from the contract; we never write an unmeasurable item in as an acceptance criterion.
| Criterion | Threshold | Measurement tool |
|---|---|---|
| AI crawler policy applied | Every agreed crawler token explicitly declared in robots.txt | robots.txt · Search Console robots.txt report |
| Content readable without JavaScript | Main content present in the first HTML response | Raw HTML response · render with JavaScript disabled |
| Structured data validity | 0 errors | Schema.org Validator · Rich Results Test |
| Entity and NAP consistency | Identical on every surface, 0 inconsistencies | Consistency audit checklist |
| llms.txt | Published and under 6 KB | File check |
| Brand citation sampling | Date-stamped record on the 25-question set, compared period over period | ChatGPT · Claude · Perplexity · Gemini · AI Overviews |
| AI crawler hit frequency | Traceable and reportable in the server logs | Server access logs |
What do we not promise?
We do not promise appearance in a specific engine, a citation frequency or a ranking position. Generative model output is probabilistic: the same question can produce different answers on the same day. What we measure is a sample rather than a census, and the report states that plainly.
What we commit to
- Measuring and recording all seven acceptance criteria
- Implementing the AI crawler access decision with its written rationale
- Zero validator errors in the structured data
- Documenting the measurement method so that it can be repeated
- A measurement date and the tool used on every reported line
What we do not commit to
- A guarantee of appearing in a particular assistant
- A guaranteed citation frequency or ranking position
- Editing engine output directly — only the source signals can be corrected
- Removing a competitor brand from the answers
- A fixed calendar date for results — the engines update on their own rhythm
Once measured values are published they appear on the proof page with the measurement date and the tool used. We do not leave a line we cannot fill marked “coming soon”.
Which companies see a concrete return from GEO?
Companies whose buyers run a long research process before purchasing. In manufacturing, export, dealer networks and holding structures, evaluation takes weeks, and the first step of that process is now often a broad question typed into an assistant. Being on that first list is decisive.
- 01 Manufacturing and factories Procurement teams building a supplier list look for capacity, certification and product-group data; when that data is not machine-readable, the company never enters the list at all.
- 02 Export industry An overseas buyer asks “which companies in Turkey manufacture this product” in their own language; without multilingual entity signals the answer returns directory sites only.
- 03 Franchise and dealer networks For questions such as “which brands grant franchises in this city”, if the central identity contradicts the location records the engine avoids using the inconsistent data as a source.
- 04 Corporate holdings When the holding and its subsidiaries are not modelled as separate entities, the engine describes the group structure incorrectly — and once a wrong fact settles in, it keeps being reproduced.
- 05 Retail chains Product, campaign and store data changes quickly; outdated sources stay in circulation inside assistants and send customers to the store with the wrong expectation.
Frequently asked questions about GEO
What is GEO?
GEO (Generative Engine Optimization) is the work of turning a website into an entity that generative AI engines use and cite while composing an answer. The target is not a ranking position; it is the appearance of the brand name and a link to the page inside the text written by engines such as ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews.
How does GEO differ from SEO and AEO?
All three share the same technical foundation and differ in target. SEO targets the ranking on the search results page, AEO targets being quoted directly in the search engine answer box, and GEO targets being named as a source inside the answer a generative engine writes.
GEO does not replace SEO. Without crawlability, indexability and content quality it does not work at all, which is why the audit always starts from the technical foundation.
Should I allow GPTBot and ClaudeBot?
This is a strategy decision and your company makes it. If you allow them, your content becomes usable as a source in generative answers; in exchange it is read and summarised. If you block them, your content cannot be used in this channel and the discoverability goes with it.
The decision is not a single switch: training, answer generation and search visibility are governed by different crawler tokens. According to Google's documentation, for example, Google-Extended affects only Gemini apps and Vertex AI grounding, not ranking in Google Search or eligibility for AI Overviews. We present both scenarios in writing and implement the decision you take.
What is llms.txt and does it actually help?
llms.txt is a proposed file published at the site root that summarises, in plain text, what the site is and which pages matter; it was published on llmstxt.org in 2024.
No major generative engine provider has confirmed in public documentation that it uses the file when selecting sources. For that reason we publish it, keep it under 6 KB and never present it on its own as a visibility promise. It costs little and carries no risk; overstating it only creates a false expectation.
How do you get into Google AI Overviews?
According to Google's documentation there is no separate markup, submission route or paid entry for AI Overviews; eligibility follows the same rules as ordinary Search eligibility.
In practice the work is this: the page has to be indexable, snippet permissions must not be switched off (nosnippet and narrow max-snippet values are reviewed), the answer to the question has to be given clearly at the top of the page, and the structured data has to be free of errors.
How are GEO results measured?
A fixed question set is run periodically with a fixed method on the same engines. On every run we record, with a date stamp, whether the brand was mentioned, which page was cited and whether the information given was accurate.
This is a sample rather than a census, and the report says so. Two supporting signals are tracked alongside it: AI crawler hit frequency in the server logs and assistant-referred traffic.
What if an AI describes my company incorrectly?
First we look for the source of the incorrect statement. In most cases it turns out to be an outdated directory record, inconsistent NAP data, a page that no longer exists, or a fact that was never written down anywhere on the site.
The correct information is then published openly, dated and marked up with structured data, and inconsistent external records are corrected. Engine output cannot be edited directly — the only thing that can be fixed is the source signals. Correction is therefore a publishing job, not a support request.
For GEO, is content volume or content quality more important?
Quality and clarity. When a generative engine writes an answer it uses the passage that contains the answer, not the whole page. Passages of 40 to 60 words that stay accurate out of context and carry a source and a date therefore outperform dozens of pages repeating the same topic.
Volume only creates value when each page answers a different question. Five pages answering the same question make it unclear which one the engine should pick, and the usual outcome is that it picks none of them.
Do you deliver GEO outside Turkey, and in which languages?
Yes. We are based in İzmit, Kocaeli and work remotely with head offices and buyers outside Turkey. GEO is delivered in Turkish and English: the entity graph, llms.txt and citable passages are built for both language versions and linked by reciprocal hreflang.
The monitoring question set is run separately per language, because assistants answer the same question differently in different languages. For exporting manufacturers this matters more than it first appears — the buyer asks in their own language, and the answer is assembled from sources in that language.
See the measured state of your website within five working days.
The audit is free and creates no obligation to work with us. The report itemises findings on AEO answerability, AI crawler access, lab-measured Core Web Vitals, structured data validity and accessibility (automated scan).
We work with corporate-scale, multi-location or multilingual organisations. One-off small jobs fall outside our scope; in that case we point you to smaller studios.
Baki Bilişim · Karabaş Mah. Salim Dervişoğlu Cad., Ncity AVM, Kat 2 (Bowlingo katı), İzmit / Kocaeli, Türkiye · +90 507 817 27 17 · bilgi@bakibilisim.com