Generative Engine Optimization (GEO) & AI Search Visibility

Generative engines first retrieve candidate passages from a search index or by fetching pages live, then synthesise an answer and attach citations. Google documents that AI Overviews and AI Mode draw on the standard Google index, that AI Mode splits a question into several sub-queries (query fan-out), that no special markup is required, and that snippet controls such as nosnippet and max-snippet apply (Google Search Central). A page that is not indexed, or whose snippet is restricted, cannot be cited there.

OpenAI documents a separate search crawler, OAI-SearchBot, which respects robots.txt and picks up changes in roughly 24 hours; a user-initiated agent, ChatGPT-User, which fetches a page when a user asks about it; and a training crawler, GPTBot, a separate token, so a site can allow search while disallowing training (OpenAI documentation). A robots.txt line that opts out of training but also blocks the search crawler removes the site from ChatGPT search entirely.

The second stage is understanding. Models resolve your brand as an entity from Organization and Person markup, sameAs-linked profiles, directories, press and industry sources. When those contradict each other, the model omits you or describes you wrongly. Entity visibility is the half of GEO that content work alone does not cover.

The third stage is citation. Engines favour passages that answer the question on their own, open with a clear definition and contain concrete names, steps or figures; definition blocks, tables and FAQ sections make them easy to extract. Google says no additional structured data is required for its AI features, but that it helps content be understood and must match the visible text (Google Search Central). We call this passage-level citability.

Our GEO consultancy begins with an audit, continues with content and entity architecture, and runs on measurement; we do not recommend GEO work without measurement.

The audit builds a panel of 50–100 prompts per target language: brand-name, category, comparison, problem-led and local questions. Each is run in ChatGPT, Perplexity, Google AI Overviews / AI Mode, Gemini and Copilot, recording mentions, citations, context (positive, neutral, wrong) and which competitors are cited instead.

We then trace causes: your top 20 pages are scored for passage-level citability, robots.txt and llms.txt are reviewed crawler by crawler, Organization, Person and Service markup and sameAs links are validated, and brand facts across the open web are compared. Every gap is tied to a cause and set into a 30-day plan with owners.

The architecture map names one responsible page per question cluster and gives it the structure the question demands: a definition block, a comparison table, a step list or an FAQ section. Existing content is mostly restructured rather than rewritten; missing topics are produced natively in Turkish, English, Arabic and Persian.

On the entity side, brand name, description, founders, locations and service list are defined once and mirrored into Organization, Person and Service markup, sameAs-linked profiles and directory listings. Third-party mentions are prioritised by source quality; no bulk directory submissions, no paid links.

Measurement re-runs the panel on a schedule and tracks share of citations (the proportion of panel answers citing you as a source), brand mention rate and position relative to competitors, reported per language and per platform, because being cited in ChatGPT and absent from Perplexity is common.

We add GA4 referral traffic from chatgpt.com, perplexity.ai, Copilot and Gemini and its conversion behaviour. Otterly.AI, Profound or Ahrefs Brand Radar can scale the panel; a small manual panel is a legitimate start.

Answer engines are not deterministic. The same prompt on the same day can cite different sources; personalisation, location and session history change the result. No agency can commit to you being cited for a given question, and we do not. We commit to a measurable method and a panel re-run the same way every quarter.

Retrieval and citation logic is not publicly documented. Google and OpenAI publish crawler behaviour and some principles, not why one passage is chosen over another. Our method rests on published documentation, observable behaviour and our own panel data; it makes no claim to have reverse-engineered a citation algorithm.

Measurement is a sample: a panel of 50–100 prompts tracks a representative slice of what buyers ask, and single-number "AI visibility scores" differ from one tool to the next, so our reports always state the panel definition, the number of runs and the date. AI-referral traffic is, for most sites today, small next to organic search, and assistants do not always show a clickable source; the effect often appears in direct traffic, brand searches and "we saw you in ChatGPT" in a sales call. Platforms change quickly, so the crawler behaviours here reflect their documentation at the time of writing.

This page follows the structure we recommend: a definition block first, then a comparison table, an FAQ and Service markup that matches the visible content. For clients we prepare a crawler policy that documents search, user-initiated and training access separately.

We put GEO on top of SEO, not in place of it: crawling and indexing from technical SEO, citable content from editorial production, entity work from digital PR, measurement from the analytics setup. We do not promise numbers for something we have not measured; we share the baseline, re-measure at quarter end and report the difference with its causes. Our work concentrates on Türkiye, the Gulf and Europe, tracking Turkish, English, Arabic and Persian questions in one panel.