GEO18 October 20267 min read

We Scanned Turkey's 100 Biggest Websites: Can AI Read Them?

People now ask ChatGPT, Claude and Perplexity "which bank", "which hospital", "where should I buy" as often as they ask Google. To answer, these assistants read websites with their own bots. We measured how ready Turkey's best-known websites are for that.

How we measured

Date: 16–17 October 2026
List: 100 well-known sites from 19 sectors (e-commerce, banking, news, airlines, hospitals, universities, automotive, food and more)
Tool: WoodstoneStudio site scan. Each homepage, robots.txt, sitemap, llms.txt and related files were checked across 10 areas and 78 signals
Bot test: live homepage requests with the identities of Googlebot, Bingbot, ChatGPT and Perplexity bots
Speed: Google PageSpeed, mobile lab measurement

Limits, stated plainly: we looked at homepages only. The scan ran from a server outside Turkey, so some sites redirected us to another country version or didn't open at all. The live bot test is a hint: firewalls may rightly reject spoofed bot identities. The numbers are a snapshot.

Finding 1: 37 of 100 sites are closed to automated reading

27 sites returned a verification page ("are you a robot?").
10 sites returned an error code or didn't respond in time.

These include some of Turkey's most visited e-commerce, classifieds and grocery sites. That doesn't make them bad sites; bot protection is a legitimate defence against fraud and load. But the bots of ChatGPT, Claude and Perplexity also come from servers abroad, and from the outside we can't tell whether these sites let them in separately. If they don't, AI answers will cite what others write about them instead of their own pages.

We left these 37 sites out of the figures below. The remaining 63 sites averaged 72.6 out of 100 (median 78).

Finding 2: the door is mostly open to bots

57 of the 63 sites (90%) allow AI search bots in robots.txt. The 6 that block them:

amazon.com.tr: ChatGPT, Claude and Perplexity search bots plus training bots
cnnturk.com: ChatGPT and Perplexity search bots plus training bots
hurriyet.com.tr: ChatGPT and Perplexity search bots plus training bots
milliyet.com.tr: ChatGPT and Perplexity search bots plus training bots
ntv.com.tr: ChatGPT and Perplexity search bots plus training bots
flo.com.tr: Claude only (search and training)

Five of them being news or media is no coincidence: publishers are deliberately pushing back on AI companies using their content. It's a choice, with a cost: a site that blocks a search bot can't be cited in that assistant's answers. Anyone who only wants to opt out of training can keep search bots open and block just the training bots (GPTBot, ClaudeBot, Google-Extended).

Silent block: on 4 sites robots.txt allowed the bots, yet the homepage returned an error or verification page to a request with the ChatGPT search bot's identity: beymen.com, setur.com.tr, turktelekom.com.tr, vakko.com. This usually comes from a CDN or firewall setting the owner may not know about. Server logs are needed to confirm it.

Finding 3: the door is open, but there's no sign

Even when AI can read a site, describing it correctly needs information machines can parse without errors.

37% (23 sites) have no structured data (JSON-LD) at all. Structured data is the machine-readable answer to "whose site is this, what do they do, where are they".
On 10 sites, organisation or brand information is barely defined (brand area score below 30).
60% (38 sites) have no llms.txt. Its impact is still limited, but it costs nothing.

Area averages make the picture clear:

AI access: 94
Technical SEO: 92
Language: 91
Security: 85
Speed: 68
Content and answers: 64
Brand and entity: 60
Structured data: 59
Share previews: 59

The technical foundation is solid. What's weak is how sites describe themselves: brand, structured data and question-and-answer content.

Finding 4: slow on phones

Across the 39 sites PageSpeed could measure:

Average mobile score: 43 out of 100
23 below 50, only 1 at 90 or above
Median time for the main content to appear (LCP): 12.5 seconds; none under Google's recommended 2.5 seconds

This is a lab measurement that simulates a slow mobile connection; real users may see faster loads. The ranking holds, though, and Google uses speed in search rankings.

Sectors

Average scores for sectors with at least three measured sites:

Home and living (3 sites): 84
Health (5): 83
Telecom (3): 79
News and media (11): 78
Automotive (4): 78
Banking and finance (9): 73
Travel (5): 71
Fashion (7): 70
Universities (6): 60
Food (3): 54

Universities stand out: none of the six measured universities has structured data.

5 steps for your own site

Check robots.txt: are search bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot, ChatGPT-User) allowed? To opt out of training, block only GPTBot, ClaudeBot and Google-Extended.
Check your CDN: on Cloudflare's AI Crawl Control use "Disallow AI training" instead of "Block" and keep search bots allowed.
Add structured data: at least Organization (or LocalBusiness), WebSite and FAQPage if you have FAQs.
Write your FAQs on the page: turn the questions customers really ask into headings with a clear 2–3 sentence answer below.
Measure mobile speed: shrink the hero image, convert it to WebP and remove unneeded scripts.

See where your site stands on these 78 signals for free: woodstonestudio.com/en/site-audit

Is it bad if a site blocks AI bots?

It's a choice. A site that blocks a search bot isn't cited in that assistant's answers. Those who only want to opt out of training can keep search bots open.

Are the numbers definitive?

No. They are a snapshot of homepages on a given date. Sites change, the live bot test is a hint, and speed is measured in lab conditions. If you spot an error, write to us and we'll correct it. Site names appear only to report technical signals measured on their public homepages.