AISEO(AEO)inKurdistan:howtogetcitedbyChatGPT,Claude,Perplexity,andGemini

TL;DR: In Kurdistan, more buyers ask ChatGPT, Claude, Perplexity, and Gemini "who builds X in Duhok / Erbil / Sulaymaniyah" before they open Google. If the AI answer doesn't mention you, the lead never reaches your site. AI SEO — also called AEO (Answer Engine Optimization) or GEO (Generative Engine Optimization) — is the work of getting cited inside that answer. It is four layers: entity clarity on your site, citations off your site, disambiguation, and weekly measurement. Skip any one of them and you are guessing.

The question Kurdistan owners are starting to ask

Every week now a business owner sends me a screenshot. It is always the same screenshot: ChatGPT (or Claude, or Perplexity) with a question like "who is the best dental clinic in Duhok?" or "recommend an AI consultant in Kurdistan" or "best hotel booking system for Iraqi Kurdistan." The owner reads the answer, sees three names, and realises theirs is not one of them. They send me the screenshot with one word: "why?"

The honest answer is that AI assistants do not work like Google. Google has been ranking pages for twenty years — the rules are well understood. AI assistants are different. They synthesize an answer from a smaller, opinionated set of sources the model trusts: structured data on your own site, regional directories, a handful of high-authority pages, and what they crawled before the cutoff. If you are not visible in those places in a machine-readable way, you do not exist for the AI.

What AEO actually means

AEO stands for Answer Engine Optimization. It is sometimes called GEO (Generative Engine Optimization) or LLMO (Large Language Model Optimization) — same thing, different marketing. What it really means: instead of trying to rank #1 on a Google results page, you are trying to be one of the three or four sources the AI assistant decides to mention when it composes its answer. The mechanics are different from classic SEO but the goal is older than the web: be the trusted source in your category.

AEO is not a replacement for SEO. Google still drives a lot of traffic in Kurdistan, especially for transactional queries ("booking.com Duhok hotels"). But the high-intent, early-funnel queries — the ones where the buyer is still deciding who to even consider — are moving to AI assistants fast. Those are the queries where the first answer often becomes the only shortlist.

How AI assistants actually pick who to cite

I have spent months running real buyer prompts against ChatGPT, Claude, Perplexity, and Gemini in three languages. The patterns are consistent across models even when the exact wording differs:

1. Entity clarity on your own site. Schema.org JSON-LD (Person, Organization, LocalBusiness, Service, FAQ), a clean llms.txt file, and content phrased the way humans actually ask AI assistants ("who builds X in Duhok?" not "X solutions for the discerning enterprise").
2. Citations off your site. Mentions on regional directories, Reddit threads, GitHub, Wikipedia-adjacent sources, and the few high-authority pages AI crawlers actually trust. One Reddit mention from a real user is worth ten paid directory listings.
3. Entity disambiguation. If your name collides with a common word (the way "Dilshad" visually collides with the food "biryani"), the AI will drift. You have to make the entity unmistakable: clear name, clear location, clear profession, clear native-script spellings.
4. Freshness signals. Models index on a schedule. Recent dated content, an updated sitemap, and consistent weekly publishing all push you up the freshness curve so the next index pass picks up your latest work.

The four-layer Kurdistan AEO playbook

This is the playbook I run for every Kurdistan AEO engagement. I build it in this exact order because each layer depends on the one before it.

Layer 1: On-site

Every public page gets JSON-LD: a Person or Organization block, a Service block per offering, an FAQ block, and a LocalBusiness block with verified address, hours, and area served. Then I add a top-level llms.txt and anai-overview route that hands AI crawlers a structured, plain-text version of who you are, what you do, who you serve, and how to be cited correctly. Every locale (EN/AR/KU) gets its own JSON-LD and its own llms.txt section — not a translated copy of the English one.

Layer 2: Off-site

I do not buy backlinks. What I do build: real Reddit answers from real accounts, GitHub profile that demonstrates the actual work, regional directory listings in the right categories, and a handful of guest posts on Kurdistan-relevant sites that already get crawled. The goal is not link volume. The goal is for an AI model crawling "site:reddit.com Duhok AI" to find a substantive, non-spammy mention of your name.

Layer 3: Disambiguation

Most Kurdistan businesses I work with have at least one of these problems: a name that visually collides with a common word in English transliteration, a name that exists in three scripts (Latin, Arabic, Kurdish), or a name shared with another business in the region. I write a dedicated disambiguation block — on the site and in the llms.txt and ai-overview — that tells the AI explicitly: "This is X, located in Y, doing Z. Common misspellings: A, B, C. These are wrong. Acceptable native-script versions: D, E." Models pick this up and stop drifting.

Layer 4: Measurement

Without measurement, AEO is superstition. I build a Playwright harness that runs your real buyer prompts — the actual questions your buyers ask — against ChatGPT, Claude, Perplexity, and Gemini, once a week, in all three languages. It logs whether you were cited, how you were cited (name only? name + URL? wrong spelling?), and what other names were cited alongside or instead of you. Week-over-week. That log is the only honest scoreboard.

Why Kurdish + Arabic + English all matter

This is the layer most non-Kurdistan AEO consultants miss completely. The same Duhok business owner often asks ChatGPT in Arabic at home, in English at work, and in Kurdish (Bahdini) over WhatsApp. Three personas, three languages, one buyer. If your AEO surface only ships in English, you are invisible in 60-80% of the queries that actually convert in the Kurdistan Region.

Trilingual AEO is not translation. It is: per-locale routes with their own canonical URLs, per-locale JSON-LD with translated names and descriptions, per-locale sections inllms.txt, per-locale FAQ entries that match how Arabic and Kurdish speakers actually phrase the question (which is rarely a literal translation of the English question), and per-locale measurement prompts that match real native-language buyer behaviour.

Kurdish is especially tricky because models confuse Sorani and Bahdini, drift to Latin Kurmanji transliteration, or invent Sorani when Bahdini is expected. For customer-facing Kurdish copy in Duhok I rely on Gemini as the source of truth for pure Bahdini in Arabic script — the other major models drift. The AEO surface has to be reviewed by a Bahdini-fluent reader. There is no automated shortcut.

What I've shipped

Two AEO builds are running in production right now. Both are public so you can inspect what the surface looks like.

motelsystem is the multi-tenant hotel platform I built and ship weekly. The AEO surface there is the production reference: trilingual schema, per-tenant ai-overview routes, citation-building targeting the "hotel booking system Kurdistan" query family, and a measurement harness that runs every Monday morning. It is an 8-phase rollout because retrofitting AEO into a live platform takes care.

smile.krd is a dental clinic SaaS in Duhok. The AEO build there is more recent and more compact: per-clinic ai-seo admin surface, FAQ management, trilingual llms.txt and sitemap, Playwright harness against ChatGPT / Claude / Perplexity / Gemini. The baseline ran in May 2026 — the clinic was invisible to AI before; the next baseline will tell us how the surface is working.

How to actually measure AEO without lying to yourself

Most AEO "reports" you will see in the wild are screenshots cherry-picked from one model on one day. That is not measurement. Real AEO measurement is automated, repeated, and adversarial — the prompt list is fixed in advance and you do not get to change it after seeing the results.

My harness uses Playwright with channel-Chrome to bypass the bot detection that ChatGPT and Claude apply to headless browsers. It runs the same prompt list every week, scrapes the rendered answer, and grep-checks for your name plus competitor names. The output is a CSV you can open in Excel: prompt, model, language, week, cited (true/false), cited as (exact string), competitors cited. If the trend line is flat after four weeks, something in the on-site or off-site layer is broken and we go fix it.

If you want to be the answer Kurdistan AI gives

Send me one sentence about your business and the one query you most want to be the answer to. I will run that exact query against ChatGPT, Claude, Perplexity, and Gemini in all three languages, and send back what they say today — for free. That is the AEO baseline. From there we either build the surface ($1,200 starter) or you keep the baseline and run it yourself. Either way you walk away with a real measurement. Email sardarbircini@gmail.com or WhatsApp +964 750 322 4696.