Every raw request retained · public counts reflect crawler purpose

Which AIactually came?

Autonomous crawls of the same URL count up to three times per half-hour; user-triggered AI fetches count per request. Neither is a user count or recommendation.

Latest 5 visit records

TypeIPTime (HK)Purpose
Machine66.220.*.*18/10 09:12—
Human204.101.*.*18/10 06:25—
Human149.56.*.*18/10 05:49—
Human106.75.*.*17/10 18:39—
Machine54.183.*.*17/10 02:59—

Data overview

Counted reads show activity; distinct URLs show coverage. Autonomous re-crawls have a half-hour cap, and raw requests remain auditable.

All-time totals

Since 2026-09-01 · machine-readable files counted from 2026-09-23
1,865readsAI readsTimes AI tools such as ChatGPT, Claude and Gemini read our pages
15kindsAI programs identifiedHow many different AI programs we could name
95URLsPages read by AIHow many different pages AI has read (merchant pages, articles…)
77devicesHuman visitorsReal people on phones or computers, counted once per device
775Visible merchant viewsHome/directory card or merchant page header at least 50% visible; counted once per device per merchant
17Merchant contactsTaps on a merchant’s WhatsApp, email or phone button
2Visits from sharesMerchant pages opened from a link a friend shared
8Merchants listedMerchants currently public in the directory

Before 2026-09-23 this site only logged crawls of its web pages. From that day on, AI fetches of llms.txt, merchants.json and the plain-text article files are counted as “AI reads” too, so the two periods are not directly comparable. Probe requests (robots.txt, sitemap.xml) and requests that match no page are never counted. Of these, 1,749 were web pages and 78 were machine-readable files; a further 255 probe requests are not counted.

Read trend

Counted AI reads / search requests (left) · first-time devices (right)

0200400600020040060007/2209/0410/18coverage expanded
AI agentsSearch crawlersFirst-time visitors

For “AI agents”, an autonomous crawler counts up to three reads of the same URL per half-hour; user-triggered fetches count per request. “Search crawlers” count raw requests. “First-time visitors” counts devices seeing a merchant for the first time that day; the same device is never counted twice. The trend only shows completed dates, so today’s partial data is not plotted.

AI crawler sources (last 90 days)

1,016meta-externalagent
375GPTBot
133Reflectionbot
96Applebot
61ClaudeBot
51KeenableBot
32ShapBot
79Other AI crawlers
One by one

AI read log

Autonomous crawlers count up to three reads per URL per half-hour; live fetches count requests. Bars compare counted records, not people.

Model trainingIndexingLive fetchUnclassified
Model training
meta-externalagentMeta1,016 counted autonomous reads
Model training
GPTBotOpenAI375 counted autonomous reads
Unclassified
ReflectionbotReflection AI133 counted autonomous reads
Indexing
ApplebotApple96 counted autonomous reads
Model training
ClaudeBotAnthropic83 counted autonomous reads
Indexing
KeenableBotKeenable51 counted autonomous reads
Indexing
ShapBotParallel32 counted autonomous reads
Unclassified
GoogleOtherGoogle30 counted autonomous reads
Indexing
PerplexityBotPerplexity17 counted autonomous reads
Live fetch
ChatGPT-UserOpenAI12 on-demand fetch requests
Indexing
AmazonbotAmazon8 counted autonomous reads
Live fetch
Google（抓取）Google6 on-demand fetch requests
Live fetch
Claude（桌面 app）Anthropic4 on-demand fetch requests
Model training
CCBotCommon Crawl1 counted autonomous reads
Live fetch
Claude-UserAnthropic1 on-demand fetch requests
01

A read does not mean indexed, quoted, or recommended.

Of the 1,865 counted records above, 518 came from an IP on that company's published crawler list. Every raw request remains available for audit.

02

Unverified does not mean fake, for two reasons. First, some companies have never published an official IP list at all (Anthropic, xAI). Second, much of this traffic is an AI reading on someone's behalf: the request leaves from that person's own computer, not from the AI company's servers, so no published list could ever match it. That is not a gap in our measurement — reads of this kind cannot be verified by design. Some AI also reads without identifying itself at all, dressed as an ordinary browser on a residential proxy; those cannot be counted. This table is incomplete, and full identification is not possible.

03

From 13 September 2026, reads we trigger ourselves while testing are tagged separately and excluded from the number above. Records from before that day cannot be separated — so the number above includes some early reads we cannot say for certain were organic rather than our own tests. We will not reclassify them after the fact.

Search engine crawl log

These are search engine crawlers — they mean your details are being indexed. Deliberately not counted in the AI reads above — being indexed and being read by AI are two different things.

BingbotMicrosoft · last 19/10 06:41
1,039 reads
PetalBotHuawei · last 19/10 05:15
831 reads
GooglebotGoogle · last 18/10 14:11
103 reads
YandexBotYandex · last 19/09 11:07
25 reads
Baiduspider百度 · last 01/10 04:11
3 reads

Which crawlers we watch for

We recognise 29 companies and 50 programs. The “logged” figure on a card is the number we have actually counted; “watching” means we have counted none yet — not that it never came.

Metaprograms: 2 · last 18/10 09:17
1,016 reads
OpenAIprograms: 3 · last 18/10 12:02
387 reads
Reflection AIprograms: 1 · last 19/10 03:48
133 reads
Appleprograms: 1 · last 19/10 02:12
96 reads
Anthropicprograms: 5 · last 19/10 07:01
88 reads
Keenableprograms: 1 · last 18/10 22:35
51 reads
Googleprograms: 4 · last 18/10 14:11 · 103 search-engine fetches
36 reads
Parallelprograms: 2 · last 14/09 20:10
32 reads
Perplexityprograms: 2 · last 18/10 16:20
17 reads
Amazonprograms: 1 · last 10/09 09:12
8 reads
Common Crawlprograms: 1 · last 16/10 17:01
1 reads
Microsoftprograms: 1 · last 19/10 06:41 · 1,039 search-engine fetches
— reads
Huaweiprograms: 1 · last 19/10 05:15 · 831 search-engine fetches
— reads
Yandexprograms: 1 · last 19/09 11:07 · 25 search-engine fetches
— reads
百度programs: 1 · last 01/10 04:11 · 3 search-engine fetches
— reads
Unidentified programsdistinct programs: 67
1,415 reads
On the watch list, none counted yet: 14 Show the list ＋
Cohereprograms: 1
Watching
Diffbotprograms: 1
Watching
DuckDuckGoprograms: 2
Watching
Firecrawlprograms: 1
Watching
Google（NotebookLM）programs: 1
Watching
Google（Vertex）programs: 1
Watching
Google（購物）programs: 1
Watching
Mistralprograms: 3
Watching
Naverprograms: 1
Watching
xAI（Grok）programs: 7
Watching
Yahooprograms: 1
Watching
You.comprograms: 1
Watching
字節跳動programs: 1
Watching
搜狗programs: 1
Watching

Companies whose programs we can count right now: 15. Two reasons it can read zero, and both are real: it genuinely has not come; or it does not use its own name — some companies read dressed as an ordinary browser, which we cannot recognise and cannot attribute to any company. The 1,415 reads under “unidentified programs” above are exactly this: we caught a program visiting, but cannot say whose it was.

We will not write “it came but we could not count it” beside a company's name — we cannot prove that. If we counted it, we print the number; if not, we print “—”.

A read is not a recommendation

From a read to a sale: which layers we can measure

An AI reading a page does not mean it will recommend you. Here are the eight steps laid out one by one, with an honest note on which ones we can measure.

  1. Crawler request

    A program fetched this page.

    We measure this
    Data source

    Logged by our server, request by request. AI crawlers, search engines, link previews and unidentified programs are counted separately, with official IP/DNS checks where available.

  2. Indexed

    A search engine added the page to its index.

    Partly measured
    Data source

    Only when a merchant connects Google Search Console, and even then we see impressions, clicks and queries rather than per-page index status. A Googlebot visit does not mean the page is indexed.

  3. Mentioned in an answer

    An AI answer names the business.

    Not measured yet
    Data source

    Not started. It needs fixed questions put to AI tools on a schedule, with the original answers saved.

  4. Linked citation

    An AI answer links to your site as a source.

    Not measured yet
    Data source

    Not started. As above, the saved answers are needed to tell whether a link was included.

  5. Recommended

    An AI suggests the business to the user.

    Not measured yet
    Data source

    Not started. Mentions, citations and recommendations will be recorded separately, never combined.

  6. Visit

    A person arrives from search or an AI tool.

    Partly measured
    Data source

    We record the referring website (domain only), plus Search Console clicks where a merchant has connected it. Many AI apps send no referrer, so this always undercounts.

  7. Enquiry

    A customer actually contacts the merchant.

    Partly measured
    Data source

    Enquiries sent through a merchant page's form, with consent to pass them on, are recorded. For WhatsApp, email or phone taps we only know the button was pressed, not whether a message was sent.

  8. Sale

    A customer actually buys.

    Not visible to us
    Data source

    Only the merchant knows. The platform cannot see it.

Each layer is a separate event. One happening does not mean the next will, and we never estimate one layer from another.

How we keep these numbers honest

  1. 01
    One device is counted once, forever

    For both views and clicks, only the first time your device sees a merchant counts. Looking again or clicking again never adds to the number.

  2. 02
    A view means it was actually on screen

    A merchant in the directory counts as viewed only when it scrolls into your screen. Loading the page alone does not count.

  3. 03
    A click means a real click out

    A click-through is counted only when someone presses a merchant's action button or link and actually goes to the merchant's own site. Clicking a merchant name to open our own merchant page does not count.

  4. 04
    Machines and humans are listed separately

    Every visit at the top of this page is labelled as either a human browser or an automated program. We do not count machines as visitors to pad the figures.

  5. 05
    Full IPs are never stored

    The full IP is masked before it is stored — we cannot, and will not, recover any visitor's full IP.

  6. 06
    Our own devices don't count

    Since 16 Oct 2026, our team's own devices are set to stay out of every view, click and visitor figure. Earlier records are kept as they were, not edited after the fact.

How we separate humans from machines

Why you see overseas IPs: Automated browsers in cloud data centres (AI browsing modes, monitoring tools, deployment checks) mostly run in the US, so their records show the US. We label all of these as machines and never dress them up as visitors. How we tell: first the User-Agent they declare; if that looks like an ordinary browser, we look up the hostname — a data centre cannot hide it (AWS answers ec2-….amazonaws.com), while home broadband does not look like that. If the hostname tells us nothing, we look at behaviour: three or more “new devices” appearing from the same network block within two minutes is not something a human does, so we count it as a machine.

Want AI to have one clear page it can read?

Create a merchant profile free. No payment, and it goes live once you confirm by email.

Create your free profile →