In English, G2 Tells AI the “Best” Tool. In Turkish, That Seat Is Empty, and Ads Fill It.

10 sectors, 2 engines, logged out. In English G2 feeds AI’s “best tool” answer. In Turkish that seat is empty, filled by ads. Zero neutral reviewers.

This week I ran a live test across 10 sectors, on two engines (ChatGPT and Perplexity), both logged out. One question: when AI answers “what is the best X,” who actually feeds the answer? You already know the English-language answer: G2, Capterra, PCMag. So I ran it in Turkish, a market with no independent-review layer, to see what AI leans on when that layer simply doesn’t exist. Across 20 searches, the number of independent, neutral local review institutions was zero. The gap gets filled by vendor listicles, commission-driven comparison sites, and, I am not kidding, a recipe site. It is a gap. It is also a wide-open opportunity. If you work in English, you take something for granted: when someone asks AI “what is the best CRM,” the answer leans on independent review institutions. G2. Capterra. PCMag. Wirecutter. These outfits test dozens of products and score them. That is their business model. You have spent years trying to get placed inside them, and you know how hard it is. Now here is a question you have probably never had to ask: what does AI lean on when that independent layer does not exist at all? That is not a hypothetical. It is the daily reality of most non-English markets. So this week I stopped theorizing and measured it, in Turkish, across 10 sectors.

The finding: 10 sectors, 2 engines, logged out. Independent local review institutions: zero

In English, AI’s “best tool” answer is fed by independent review institutions. In Turkish there is no such institution. Across all 10 sectors, on both engines, that seat was empty, and the gap was filled by advertising, commissions, and SEO filler.

Method (let’s be transparent, this is the whole point)

One question: who feeds AI’s “best X” answer, and does the picture change between English and Turkish? Sample: 10 sectors, deliberately mixed: CRM, email marketing, hosting, VPN, robot vacuum, washing machine, credit card, private health insurance, online English course, accounting software. B2B software all the way to white goods. Two engines, both logged out: ChatGPT (guest, web search on) and Perplexity (signed out). I stress logged out on purpose: it removes personalization and account history, so anyone can reproduce it. One technical note: with web search off, ChatGPT shows no sources at all, it answers from memory. I had to switch search on to see citations. That itself confirms last week’s mechanism: search off is frozen memory, search on is live grounding. Limits (honestly): one day, one session, one query per sector. Source classification is my judgment. This is not a verdict on any sector, it is a reproducible probe. And a probe like this should be run regularly, not once. More on that at the end.

The table: who fed the answer in 10 sectors?

Sector	ChatGPT (logged out, web search)	Perplexity (logged out)	Independent local institution?
CRM	G2, Forbes, TechRadar (foreign)	Vendor listicles	None
Email marketing	TechRadar, Software Advice	Vendor + foreign media	None
Hosting	TechRadar (foreign)	Affiliate/SEO blogs	None
VPN	Tom’s Guide, RTINGS (foreign)	Foreign affiliate + YouTube	None
Robot vacuum	Tech media + retailers	A recipe site (!)	None
Washing machine	Tech media + retailers	A celebrity-gossip site (!)	None
Credit card	Commission-driven aggregators	Aggregators + forum	None
Private health insurance	Insurance brokers	Brokers + brand pages	None
Online English course	Foreign blog	Affiliate blogs + YouTube	None
Accounting software	Accountant + vendor blogs	Vendor listicles	None

Headline number: 10 sectors x 2 engines = 20 live searches. Independent, neutral local review institutions in the citations: zero.

Three lessons

Lesson 1: In English there is a jury. In Turkish the jury seat is empty. In English, G2, Capterra, and PCMag test many products and rank them. In Turkish, not one such institution appeared, in any of the 10 sectors. Lesson 2: Two engines, two escape routes, same void. For a Turkish query, ChatGPT jumps straight to foreign English institutions (G2, Forbes, TechRadar); the local vendors do not even appear. Perplexity stays in Turkish but takes whatever it can find: vendor listicles, and in one case a recipe site as the source for “best robot vacuum,” a gossip site for “best washing machine.” One borrows a jury from abroad, the other asks a random passerby. Lesson 3: AI takes the easy path because it has no choice. The model is not lazy, it is stuck. With no neutral local address to go to, the gap is filled by ads, commissions, and SEO content. In English the dynamic is different: a strong independent-review ecosystem exists, and the model leans on it. In Turkish that layer was never built. This is the pattern in every emerging market, and it is exactly where the opportunity hides.

What this means for you

This looks like a limitation. I see an empty chair. In English, claiming that chair is close to impossible; replacing G2 or Capterra takes years and millions. In a market without that layer, the chair is empty, and today it is occupied by mediocre, self-serving lists. Whoever fills it with genuinely neutral, test-driven, evidence-backed content becomes the natural reference of the AI era. That is an opportunity for content teams and, frankly, for founders and investors: build the “G2” of one vertical in an underserved market and the upside is large. Concretely, run the test in your own space (incognito, logged out, both engines, look at the sources), then write the content that fills the gap the way Peec AI describes the ideal citation chunk:

• 2 to 3 sentences, around 50 words.
• Self-contained, makes sense on its own.
• Entity dense: name specific brands, models, criteria.
• Declarative and authoritative.
• Loaded with E-E-A-T.
• And it mentions your brand favorably, with proof, not a boast.

Not “this is a good tool,” but “in independent testing and 200+ user reviews, X delivered the best result in category Y.” AI puts evidence into the answer, not self-praise.

Quick hits

• Google can leave low-value AI content out of the index entirely. Chris Long flagged it: in a recent Martin Splitt and John Mueller podcast, Google confirmed that clearly AI-generated content that adds nothing unique can get a “crawled, currently not indexed” status. It reinforces this whole issue: what gets indexed, and what gets cited, is content that adds proof, not another empty “top 10.”
• The tooling rule of thumb: Peec, Otterly, and Profound run each prompt daily and report weekly or monthly averages; they never treat a single run as a finding. Same lesson for us: one day is a snapshot, regular repetition is a finding.

Closing

One sentence: in a market without an independent-review layer, no one gets a neutral answer from AI, because the institution that would produce that neutral source has not been built yet. Here is what stays with me. This looks like a deficiency, but it is an invitation. The independent-review ecosystem AI relies on in English was not built in a day; someone saw the gap and filled it. In underserved markets that gap is wide open right now. Fill it with honest, test-driven, evidence-first content and you become the reference the AI era keeps citing. A note to myself too: I ran this once, but once is not enough. Answers shift daily. I am turning this into a monthly series, an “AI Source Gap Index” for underserved markets. Open question for you: in your market and your sector, does a neutral, non-commercial review institution actually exist? Name it in the comments. If nothing comes to mind, maybe you are the one who should build that chair.