Independent AI-shopping research

We measure what AI recommends to buy.

Shopping is moving from search boxes to AI assistants. AIBUY asks 8 assistants the same 100 purchase-intent prompts per category, logs every answer, and publishes the recommendation data — so buyers get evidence and merchants get a scoreboard. No paid placements. Ever.

Verified by humans.Every report carries a card showing how many prompts × models were tested, when, and exactly which parts a person checked. Automated process — never automated judgment.
Independent  ·  Method-first  ·  Data in the open  ·  Est. 2026
Best mechanical keyboard under $100 for a programmer?” — same prompt, 8 assistants
ChatGPT6 brands
Claude5 brands
Gemini4 brands
Perplexity4 brands
Doubao3 brands
Kimi3 brands
40 prompts · Sep 2026 · full log in report Human-verified ✔
8AI assistants tracked
100+prompts per category
4public metrics
0paid placements
Why this matters

The shelf just moved. Nobody is auditing it.

For twenty years, merchants fought for Google rank. Now their customers ask an assistant — and the assistant answers from a mix of training data, third-party lists and crawlable pages. If AI misstates your specs, or skips you entirely, no one tells you. We tell you, with data.

01

Measure

One hundred purchase-intent prompts per category — the way real buyers actually ask — run through every major assistant, on a schedule, with full answer logging.

Our method →
02

Compare

Mention rate, rank position, info accuracy and competitive suppression — four metrics, published with prompt-level evidence. Same method in the US and China markets.

Cross-market project →
03

Trust

We are the auditors, not the shelf. Rankings on AIBUY cannot be bought — that independence is what makes the data worth reading and the service worth paying for.

Independence policy →
The human-verification badge

Automated process, never automated judgment

Data collection, testing, charts and first drafts run automatically. Conclusions, opinions and anything we publish as a recommendation are written and checked by a person. Each report opens with a card like this:

HUMAN-VERIFIED · AIBUY-VB-1.0
Test set100 prompts × 8 assistants
WindowSep 05–12, 2026
AutomatedData · charts · draft
HumanConclusions · analysis
Method v1.0 · prompt log included
Reports

Pilot: mechanical keyboards, US market

The full pilot — 40 prompts × 8 assistants — is being published with its complete prompt log and answer excerpts. Headline numbers from the dry run:

62%prompts where one brand led every assistant
31%of assistant answers contained a spec error
2.4×mention gap between #1 and #5 brand
17%of recommended products were outdated models
AssistantMention rate (top-3)Avg. first positionSpec accuracyOverlap with ChatGPT
ChatGPT78%1.691%
Claude64%2.188%71%
Gemini58%2.482%66%
Perplexity52%2.886%61%
Doubao44%3.374%38%
Kimi40%3.579%35%

Pilot dry-run numbers, shown to illustrate the report format. Final published tables carry the full prompt log, dates and model versions. Full report →

The cross-market project

Same question. Two AI worlds. How far apart are the answers?

AIBUY runs the identical purchase-intent prompts against the US assistant set (ChatGPT, Claude, Gemini, Perplexity) and the China set (Doubao, Yuanbao, Kimi, Wenxin, Taobao Ask). Only a team operating in both markets can produce this comparison — and we publish it quarterly, free, in English and Chinese.

“Best $300 gift for a coffee-loving girlfriend?”
US set top overlap34%
CN set top overlap41%
See the project →
For merchants

Find out what AI tells your customers about you

You optimized for search for a decade. Your buyers now ask assistants directly. Our diagnostic shows exactly where you stand — and the managed service keeps you standing there.

One-off

Agent Readiness Diagnostic

100 purchase-intent prompts × 8 assistants, focused on your category. You get your AI Recommendation Share, where you rank, which specs the assistants get wrong, and who is winning your customers' questions — with the prompt-level evidence.

What's inside →
Monthly

Managed AEO Monitoring

Monthly re-tests and a trend dashboard: share movements, new entrants, spec errors, and the visibility fixes that matter — structured data, crawl access, and presence in the third-party sources assistants actually quote.

How it works →
The red line: AIBUY rankings and reports never accept payment for placement. Merchants pay for measurement, monitoring and visibility engineering elsewhere — never for a position on our charts. Read the policy →
Newsletter

The AI Recommendation Monitor

One email a week: what the assistants recommended, what changed, and one dataset worth your time. Free, no algorithm between us — unsubscribe anytime.

Weekly · English · sample issue in the first send. 中文版请订阅 aibuy.org.cn

FAQ

Short answers first

Do brands pay to appear in your reports?

No. Reports are funded by our diagnostic and monitoring services for merchants, and by newsletter sponsorships that are always labeled. Ranking positions are computed from test data only.

Which assistants do you test?

US set: ChatGPT, Claude, Gemini, Perplexity. China set: Doubao, Tencent Yuanbao, Kimi, Wenxin, Taobao Ask. The cross-market project queries both sets with identical prompts.

How is this different from an affiliate review site?

We publish the prompts, the raw answer log, and the counting rules. You can disagree with us — but you can check us. Affiliate links, where present, are labeled and never influence scoring.

Can I get my brand tested?

Yes — the diagnostic is a one-off deliverable, and we run a limited free pilot each quarter for categories we're covering. See services →