Most businesses are now invisible to the search engines their customers actually use, and the reason is embarrassingly fixable. On October 12, 2026, Stradiji, an SEO and generative-engine-optimization consultancy based in Istanbul, released an AI visibility audit of more than 100 websites across seven countries, and the headline finding is stark: nearly nine in ten of the audited sites had closed at least one of the three conditions that decide whether AI engines can reach, recognize, and believe a business.
In other words, while companies debate elaborate AI content strategies, most have quietly locked the front door, torn off the nameplate, or left no references behind. Here is what the audit found, how those failures compound, and the five-minute fix that should be the first item on any marketing agenda.
The three doors of AI visibility
Stradiji's GEO Score Card framework groups 70 checks into three stages the company calls Reach, Recognize, and Believe. The names do the explaining:
- Reach measures whether AI engines such as ChatGPT, Gemini, and Perplexity can access the site at all.
- Recognize measures whether those engines identify the brand as a distinct entity.
- Believe measures whether independent sources support what the business says about itself.
The audit crawled each site in full, reviewed robots.txt, llms.txt, ai.txt, and AGENTS.md files, and ran live tests in ChatGPT, Perplexity, Gemini, and Google's AI features. The sites, based in Turkey, the United States, the United Kingdom, Germany, Estonia, Romania, and South Korea, span health, e-commerce, B2B software, real estate, media, law, manufacturing, and travel. One honest caveat: this is a practitioner sample built through client and prospect engagements through September 2026, not a random draw. Still, a near-total failure rate across 100+ sites is a signal no business should dismiss.
Door one: most sites block the very engines they want citations from
About three in four of the audited sites give AI crawlers no explicit access in robots.txt, and some actively block GPTBot, ClaudeBot, or PerplexityBot outright. Stradiji's founder Mert Erkal put the irony perfectly: "Many companies are paying to be visible in AI answers while their own robots.txt tells those same systems to stay out." He adds that most of the issues the study found cost nothing to fix, and the first one takes about five minutes.
This is where our own position is worth stating directly: this site's robots.txt is deliberately open to all major search and AI crawlers, including OAI-SearchBot, because citations in ChatGPT and its peers are now a real traffic channel, one that recent research shows compounds. Being crawlable is not a guarantee of being cited, but blocking it is a guarantee of not being.
The failure is not always deliberate. The audit flags an infrastructure shift every site owner should know about: since September 15, 2026, Cloudflare's Block setting also applies to mixed-use crawlers such as Googlebot, Bingbot, and Applebot, according to Cloudflare's own blog post that day, and new ad-supported domains get a preset that blocks AI agents on pages with ads. Companies can now block search engines and AI crawlers without intending to. Check your Cloudflare settings today, not after traffic falls.
Door two: most brands are not yet a recognizable entity
Reach is worthless if the engine that reaches you cannot tell who you are. The audit's findings on entity signals:
| Entity signal | Share of audited sites missing it |
|---|---|
| Wikidata entry | ~87% |
| Wikipedia article | ~3 in 4 |
| Basic Organization schema markup | ~30% |
This is the unglamorous plumbing of AI visibility. Wikidata, a proper knowledge-graph presence, and basic structured data are how a brand stops being "some site that mentioned plumbing" and becomes "the entity Ember & Oak Plumbing, located in Hamilton,Ontario, specializing in emergency repair." AI engines lean on structured identifiers more heavily than human readers ever did, and roughly a third of the audited sites lacked even Organization schema.
Door three: half the sites give engines nothing to believe
Roughly half of the audited sites had no independent reputation signal at all, which the study defines as a third-party review, award, or industry mention that the company did not publish itself. AI answer engines are explicit about preferring corroborated claims; a business that only exists in its own words invites either omission or hedged ("claims to...") answers. This is why visibility work and authority work are the same job now: the citations that make an answer engine trust you are the same ones that make a customer trust you.
Why fixing the doors compounds
Each closed door is cheap to open, and the payoff is not linear. Recent GEO research on 9,471 prompts found that sites cited on one topic get cited more on others across all five major answer engines: topical authority compounds. Closing doors forfeits the compounding; opening them starts it. Ranking #1 in Google, meanwhile, does not guarantee AI citations, as the AI Shelf Index showed, which means traditional SEO excellence cannot bail you out of a locked robots.txt.
The practical sequence is short:
- Today (5 minutes): Open robots.txt to major AI crawlers, and verify your Cloudflare settings did not block Googlebot and friends by accident.
- This week: Deploy or repair Organization schema, and make sure your entity data (name, sameAs links to your social and Wikidata profiles) is accurate everywhere it appears.
- This month: Seed Wikidata, chase genuine third-party mentions, and make sure reviews land on platforms engines actually read.
- Ongoing: Run live queries in ChatGPT, Perplexity, and Gemini about your own products and city, and write down what the engines say. That is your actual visibility dashboard, and it is free.
Conclusion
Stradiji's audit of 100+ sites across seven countries lands the same verdict as every serious GEO study this year: nine in ten businesses have locked at least one of the three doors of AI visibility, most commonly by blocking the crawlers of the very engines they hope will cite them, sometimes without even knowing it. The fixes are unsettlingly cheap per Erkal: access first at five minutes, then entity plumbing, then real third-party credibility. The compounding math cuts both ways, and the earlier the doors are open, the longer the advantage accrues. This is precisely the audit our team runs for clients; if you want to know which of your three doors is shut, send us a note on our contact page and we will check all 70 checkpoints against your site and hand you the fix list.
Read next:
- SolarWinds Avior: 300,000 customers get an AI agent, zero outcome data
- The EU opens its first AI Act enforcement case against OpenAI and Anthropic
- Shopify opens its catalog to any personal AI agent, with a human sign-off
- Plus tôt : The AI Shelf Index: why ranking #1 in Google won't get you cited by AI
