Chapter 3 Measurement Sources and the AI Black Box: The Data Foundation of Performance Measurement · IDAEO: Marketing in the Age of AI
Measurement Sources and the AI Black Box: The Data Foundation of Performance Measurement
This chapter answersWho provides the numbers for each step, and how is the AI black box measured?
Why it comes hereSort out source, scope, and unit first; only then can you make sense of the results.
Figure 3-1 IDAEO Measurement Service Blueprint
View as text
Read from top to bottom, from website to person, and compare each layer's results with the scope of ongoing tracking.
Service blueprint: our website → crawlers → AI Platforms → people (2026/3/3–10/8) Horizontal: the steps of this book; vertical: the service blueprint; gray band = Black Box of the AI Platforms. Numbers come from our site records and search platform data respectively. Solid + number = measured Dashed, gray = platform internal Gray band = Black Box Blue background = our records Interaction Visibility Internal interaction ▲ Measured ▼ Platform internal Humans Ask→see the answer→Organic Traffic Measured data AI front stage AI answers, search results Measured data AI back stage Training, indexing, ranking In-platform Crawlers visitingour website Requests by function:our site records Measured Our website Whole-site requests Measured Crawlers visit on their own and fetch content Ask AI Live retrieval Tasks AIAgents Auto-browsing Searches,sees results Google/Bing Sees theAI answer Cited us Organic Traffic 309,826 times Visit Gets a query Chooses to fetch Gets the task Sends AIAgents out Search impressions ~8.22M times · approx. Google+Bing AI Answer 31,711 times·Copilot Other platforms Tracking Index Are we in? In-platform Indexed pages 657K pages Google 282,000 Bing 375,000 Ranking What rank? In-platform Trains models Model usebeing tracked Builds answer,picks citations Which part ofours was used? In-platform Crawl 123.4 million Requests incl. browsing AI total Copies to train 52.2 million Requests Anthropic 24.48 million Builds AI index 3.99M requests Builds search index 41.26M requests Apple 24.62 million Live Retrieval 651,000 Requests OpenAI 635,000 Act for Users 18.09 million Requests AI Agent incl. browsing Organic Traffic 309,826 site visits S1 Prepare 196.7 million Site total 196,702,526 requests (2026/2/12–10/8); Four AI Audiences 105,487,618 + automated browsing 17,948,162, listed separately. Remaining traffic 73,266,746 requests; latest classification: Apple → search indexing, Meta → training. 1. Period 2026/3/3–10/8; AI Agents (automated browsing) 2026/8/26–10/8. Each box shows requests, pages, impressions, or Organic Traffic; the units differ. 2. The gray band is the Black Box: training, indexing, and citation passage selection happen inside the platforms; indexed pages are the September 2026 figures, Bing 375,000 + Google 282,000. 3. Search impressions about 8.22 million (Google + Bing); AI answer citations 31,711 (Copilot, 2026/7/22–10/8). 4. Organic Traffic 309,826 visits; Google 267,734, Microsoft 20,053, Yahoo Japan 10,278, Meta 9,388. 5. Search indexing and AI Overviews come from the same Search Engine Crawlers (41,261,941 requests); reads for AI answers 46,052,576 requests. S1 (requests) S2 (requests) S3 (requests) S4 (requests/pages) S5 (requests) S6 (times) S7 (times) S8 (requests) S9 (clicks) S10 (Tracking) S10 Return Tracking Indexed Model use (Tracking) Fetched in real time ID EO Prepare Crawl Copies to train Index Ask AI AI Mention AI Answer Acts for customers Visit Return
How to readRead from top to bottom, from website to person, and compare each layer's results with the scope of ongoing tracking.
Figure 3-2 Content Journey Map and the AI Black Box
View as text
Read the journey phases horizontally, comparing the number cells with the live-test topics inside the black box.
Content Journey Map for the AI Era: S1 to S10 (2026/3/3–10/8) Four phases, ten steps; gray dashed boxes mark the inside of AI Platforms. Site-wide 196,702,526 requests. Readiness Phase One · S1 Phase Two: AI Information Acquisition S2–S4 Phase Three: AI-Assisted Decision and Action S5–S8 Phase Four: Convert & Return S9–S10 S1 Prepare (requests) S2 Crawl (requests) S3 Train (requests) S4 Index (requests) S5 Ask (requests) S6 Mention (impressions) S7 Answer (citations) S8 Act (requests) S9 Visit (Organic Traffic) S10 Return (Tracking) Who acts Target Measured count Black Box Platforminternal Site share requests Volume curve square-root scale Merchant Website Search Engine AI Crawlers AI Crawlers Search Engine AI Platforms AI Crawlers Search engines AI Platforms AI Agent Humans Humans AI Platforms 196,702,526 Requests 2/12–10/8 105,487,618 Requests AI and search engines 52,203,119 Requests Training collection 45,253,224 Requests Both indexes 650,528 Requests Live Retrieval ~8.22M times Search impressions 31,711 times Copilot citations 7/22–10/8 17,948,162 Requests 8/26–10/8 309,826 times Organic Traffic Tracking Tracking Found by AIs Identity: Tracking How content is filed and ranked after reading Platform internals: Tracking How choices are made after AI mentions and answers External records: ongoing tracking Whether customers ask again later Tracking Site baseline 53.63% 26.54% 23.01% 0.33% Different units Different units 9.12% Separate log Tracking S3 and S4 are binned by each family's primary use and are mutually exclusive in this table; S6, S7, and S9 use units other than requests. 196,702,526 Phase 1 105,487,618 Phase 2 17,948,162 Phase 3 309,826 Phase 4 Source: crawler requests follow the approved list; site-wide totals, use categories, Organic Traffic, citations, and search impressions follow approved values. S4 total = AI indexing 3,991,283 + search indexing/AI Overviews 41,261,941; each type is counted once in this table. The curve only indicates the order of magnitude of the four phases: S1 site-wide, S2 AI requests, S8 automated browsing, S9 Organic Traffic; each unit shows its own magnitude. ID EO
How to readRead the journey phases horizontally, comparing the number cells with the live-test topics inside the black box.
Figure 3-3 Journey Outcome Matrix Across AI Platforms
View as text
One row per platform, S1 to S10 across; nodes mark published results, and the rest should be read as still being tracked. In the figure, S6 lists search impressions (not AI mentions), meaning the number of times you appear in search results. The mention rate in AI answers will be published later.
Journey Outcome Matrix Across AI Platforms How far each one gets (S1→S10) S1 Prepare 196,702,526 requests S2 Crawl 105,487,618 requests S3 Copies to train 52,203,119 requests S4 Index 45,253,224 requests S5 Someone asks AI 650,528 requests S6 AI Mention about 8.22 million times S7 AI Answer 31,711 times S8 Acts for customers 18,091,884 requests S9 Visit 309,826 clicks S10 Return Tracking in Progress Readiness AI Information Acquisition AI-Assisted Decision and Action Conversion & Return S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 AI Platforms/Categories Left column total 105,487,618 requests; automated browsing counted separately. Apple 24,615,429 24,615,429 Anthropic 24,492,504 24,479,305 10,705 2,494 Microsoft 12,050,061 12,050,061 375,000 pages 910,622 31,711 20,053 OpenAI 11,998,898 10,602,782 760,806 635,310 Meta 11,583,411 11,583,411 9,388 Amazon 5,454,011 5,454,011 Google 3,413,957 3,413,483 282,000 pages 474 7,410,298 267,734 Our AI Search 2,815,693 2,815,693 Yandex 511,159 511,159 ByteDance 255,863 255,863 Sogou 253,223 253,223 Perplexity 228,656 218,062 10,594 TollBit 126,292 126,292 Parallel 125,562 125,562 Huawei 114,316 114,316 Linkup 60,455 60,455 Common Crawl 19,054 19,054 DuckDuckGo 14,008 12,384 1,624 1,049 DeepSeek Merged into Tracking in Progress mcpbeat 7,925 7,925 Other platforms 115,218 64,556 36,023 32 14,607 11,602 AI Agents (browsing) 17,948,162 17,948,162 AI Tools 7,079,310 Tracking in Progress 152,613 Node height uses a log scale; gray bars represent pages, counts, or clicks. Category rows are already included in the left-side total traffic. Site-wide requests from 2/12; crawler requests from 3/3. The search impressions overview uses the approved rounded figure; per-platform breakdowns use actual data. ID EO
How to readOne row per platform, S1 to S10 across; nodes mark published results, and the rest should be read as still being tracked. In the figure, S6 lists search impressions (not AI mentions), meaning the number of times you appear in search results. The mention rate in AI answers will be published later.
OpenAI: One Platform, Three Purposes for Reading
During Mar 3, 2026–Oct 8, OpenAI's three identified crawler types made a total of 11,998,898 requests to the site: 10,602,782 for training, 760,806 for indexing, and 635,310 for real-time retrieval. OpenAI had the most real-time retrievals of any platform. These are requests recorded by the website. They do not establish that content entered a model, was indexed, or was mentioned or cited in an answer. OpenAI's later-stage results remain Tracking in Progress. See the full chart in Figure 3-7 under “Platform Profiles” at the end of this chapter.
Figure 3-4 Google: 7,410,298 search impressions｜The platform with the most complete numbers along the whole path
View as text
Read each step from left to right; check source, period, and unit first, then compare numbers on the same basis.
Google: search impressions reached 7,410,298 Finding: from indexing and impressions to customer visits, Google has the most complete numbers along the whole path. The Google path: from S1 Prepare to S10 Return 2026/3/3–10/8 · site-wide counted from 2/12 S1 Prepare S2 Crawl S3 Train S4 Index S5 Ask AI S6 AI Mention S7 AI Answer S8 Act S9 Visit S10 Return S1 Prepare Our website 196,702,526 Site requests · 2/12–10/8 S2 Crawl All Google crawlers 3,556,335 3/3–10/8 · requests S3 Train The Google path Tracking S4 Index Google search crawler 3,413,483 Requests ~282,000 pages indexed S5 Ask User-triggered 474 requests · both types S6 Mention Google impressions 7,410,298 impressions (search) S9 Visit GoogleOrganic Traffic 267,734 visits · separate log Numbers at different steps come from different records; requests, pages, impressions, and Organic Traffic are each read in their own units. Google requests Search crawler 3,413,483 Tracking 142,378 Real-time retrievals 474 Total 3,556,335 From search platform Indexed about 282,000 pages Search impressions 7,410,298 times Search clicks 179,618 times Search CTR 2.42% Organic Traffic (separate) Brought by Google 267,734 visits All sources 309,826 visits Organic Traffic and search clicks each read in its own unit Black Box: inside AI Platforms AI Answer S7 AI citations: Tracking in Progress; due later Act S8 Act: Tracking in Progress S10 Return Tracking Three records, one segment each Website requests Daily crawler request log Search platforms Indexing, impressions, clicks Organic Traffic Visitor sources visible Weekly search click-through rate (Google search platform · search clicks ÷ search impressions) 2026/3/3–10/8 · weeks start Monday · domain properties of the old and new sites 0% 2% 4% 6% 8% 3/2 3/30 4/27 5/25 6/22 7/20 8/17 9/14 10/5 7.03% 1.25% 2.63% Same period 179,618 search clicks 7,410,298 search impressions Overall click-through rate 2.42% Sources: crawler requests from the approved list; site-wide and Organic Traffic from approved figures. Indexed pages are the September search platform snapshot. Google search impressions, clicks, and weekly curve: two domain properties on the search platform, using web search records from 3/3–10/8. Search result impressions and Organic Traffic each have their own basis; Google's AI answer citations, actions, and return visits are still being tracked. ID EO
| Item | Value | Unit |
|---|---|---|
| Search crawler | 3,413,483 | Requests |
| Tracking | 142,378 | Requests |
| Real-time retrievals | 474 | Requests |
How to readRead each step from left to right; check source, period, and unit first, then compare numbers on the same basis.
Figure 3-5 Microsoft: 12,050,061 search crawler reads｜The platform that currently provides AI answer citation numbers
View as text
Compare with Figure 3-4 (Google), and note Microsoft's AI answer citation volume and its reporting period.
Microsoft: 12,050,061 search crawler reads Finding: the platform that currently provides AI answer citation numbers; Copilot cited 31,711 times. The Microsoft (Bing, Copilot) path: from S1 to S10 2026/3/3–10/8 · site-wide counted from 2/12 S1 Prepare S2 Crawl S3 Train S4 Index S5 Ask AI S6 AI Mention S7 AI Answer S8 Act S9 Visit S10 Return S1 Prepare Our website 196,702,526 Site requests · 2/12–10/8 S2 Crawl Bing search crawler 12,050,061 3/3–10/8 · requests S3 Train Training use Tracking S4 Index Bing search crawler 12,050,061 Requests ~375,000 pages indexed S5 Ask Real-time retrieval Tracking S6 Mention Bing impressions 910,622 impressions (search) S9 Visit MicrosoftOrganic Traffic 20,053 visits · separate log Numbers at different steps come from different records; requests, pages, impressions, citations, and Organic Traffic are each read in their own units. ID EO Microsoft requests Search crawler 12,050,061 Builds search index Also serves AI Overviews Same batch of requests Counted once Total 12,050,061 From search platform Indexed About 375,000 pages Search impressions 910,622 times Search clicks 32,052 times Click-through rate 3.52% Organic Traffic (separate) Microsoft brings 20,053 visits All sources 309,826 visits Organic Traffic and search clicks each read in its own unit Answers visible; actions and returns still tracked S7 Copilot answer citations 31,711 times S8 Act Tracking S10 Return Tracking Three records, one segment each Website requests Daily crawler request log Search platforms Indexing, impressions, clicks Answer monitoring Citation counts visible Weekly search click-through rate (search clicks ÷ search impressions) Bing search platform · data available 7/21–10/6 · weeks start Monday Higher Lower 7/20 8/3 8/17 8/31 9/14 9/28 10/5 Same period 32,052 search clicks 910,622 search impressions Overall click-through rate 3.52% Sources: crawler requests from the approved list; site-wide, indexing, Organic Traffic, and answer citations from approved figures. Search impressions, clicks, and the weekly curve use the daily data available from the search platform, ending 10/6; the answer citation period is 7/22–10/8. Bing indexing is the September snapshot; answer citations and search clicks are counted separately, and platform actions and return visits are still being tracked.
How to readCompare with Figure 3-4 (Google), and note Microsoft's AI answer citation volume and its reporting period.
Figure 3-6 How Do You Know AI Cites You? Measuring Both Ends of the Black Box
View as text
On the left are first-party records: AI Real-Time Retrievals of 651,000. On the right are platform data and direct questions. Google's generative AI impression figures for this site are still pending. Copilot answer citations of the IDAEO platform total 31,711. TK Lin's personal IDA appeared in AI search source lists at a rate of 84% (four of the five models used the same search interface; the questions named the author himself; a single case). Below, reads, impressions, and citations form a checkable path. Requests, impressions, citations, and question ratios use different units and cannot be added together. This is a method diagram; cross-verification has not yet been completed.
ID EO How do you know AI cites you? Cross-checking content use from both ends of the black box Seeing content use Seeing impressions, citations, answers Site / edge first-party records Platform data / asking AI directly AI Black Box Training · ranking Mention · answer · action S6 Mention S7 Answer 1 VPS access logs Site requests · paths 2 Cloudflare AI crawler requests AI referrals Matched against VPS Identifies the same read First-party total 651,000 AI real-time retrievals 3 Google GSC Generative AI impressions and the URLs shown Our site's figures pending 4 Bing Webmaster Tools Copilot answer citations 31,711 times 2026/7/22–10/8 5 Asking multiple AIs 100 questions · five models 84% IDA used as a source Same URL · same period Links evidence from both ends into one checkable path Crawls Exposure Citation / answer Requests, impressions, citations, question ratios: different units, not additive. Bing cited URLs and tested answers check citation directly; front-end records fill in reads and impressions. Case before vs. after TK Lin's personal IDA, live test 9/20 0 10/3 84% 100 questions, five AI models; 84% of answers used this IDA as a source. IDAEO measures both ends of the black box, then uses platform dataand periodic live tests to measure the results of content use. Original framework © 2026
How to readOn the left are first-party records: AI Real-Time Retrievals of 651,000. On the right are platform data and direct questions. Google's generative AI impression figures for this site are still pending. Copilot answer citations of the IDAEO platform total 31,711. TK Lin's personal IDA appeared in AI search source lists at a rate of 84% (four of the five models used the same search interface; the questions named the author himself; a single case). Below, reads, impressions, and citations form a checkable path. Requests, impressions, citations, and question ratios use different units and cannot be added together. This is a method diagram; cross-verification has not yet been completed.
Platform Profiles: OpenAI, Anthropic, Apple, Meta
All four currently have read records. Platform impressions, citations, or Organic Traffic breakdowns remain Tracking in Progress, so these profiles appear at the end of the chapter. Once the data is complete, they will return to the main text under the same rule: records must cover reads, platform impressions or citations, and an Organic Traffic breakdown.
Figure 3-7 OpenAI: 10,602,782 reads for training purposes｜Nearly nine in ten requests are for training purposes; also the platform with the most real-time retrievals
View as text
Compare with Figure 3-4 (Google): look first at retrievals and customer arrivals, then at the scope of ongoing tracking in the middle.
OpenAI: Of the three crawler types, the one for training purposes recorded the most reads: 10,602,782 Finding: Nearly nine in ten reads are for training purposes. It is also the platform with the most real-time retrievals. The OpenAI path: crawlers come in, black box inside the platform, customers visit 2026/3/3–10/8 · three crawlers, 11,998,898 requests total; OpenAI Organic Traffic breakdown still being tracked Requests from three identified crawlers Green band widths are allocated by 11,998,898 requests; each purpose is counted once. S3 Train 10,602,782 requests S4 Index 760,806 requests S5 Ask 635,310 requests Black box inside the platform AI mention, answer, and action flow OpenAI-specific requests: Tracking in Progress S1 Prepare Our website S2 Crawl OpenAI's three crawlers S3–S5 Entering the site Site request log S6–S8 Inside the platform Black Box S9 Visit Organic Traffic Whole site 196,702,526 Requests · 2/12–10/8 OpenAI 11,998,898 Requests S3 Train Training crawler 10,602,782 Requests Takes content back for model training S4 Index Search crawler 760,806 S5 Ask Real-time retrieval crawler 635,310 Training requests visible In AI Platforms: live tests Indexing requests visible In AI Platforms: live tests Real-time retrievals visible In AI Platforms: live tests After training In AI Platforms: live tests S6 Mention In AI Platforms: live tests S7 Answer In AI Platforms: live tests S9 Visit Organic Traffic Tracking Site-wide total 309,826 visits S8 Act Tracking in Progress S10 Return Tracking in Progress On this path, what we can count directly is crawler requests arriving at the site. The training, indexing, and question request branches do not overlap; they sum to 11,998,898. The platform's mentions, answers, and actions are each read in their own units. The 309,826 in S9 is Organic Traffic from all sources; the OpenAI breakdown is still being tracked. Sources: approved site-wide and Organic Traffic figures; the three crawlers' requests are summed from the 2026/3/3–10/8 approved list. Green band widths are allocated by the request counts of the three crawlers; the dashed box marks the inside of the OpenAI platform, observed by live tests. ID EO
How to readCompare with Figure 3-4 (Google): look first at retrievals and customer arrivals, then at the scope of ongoing tracking in the middle.
Figure 3-8 Anthropic: 24,479,305 requests for training purposes｜Highest number of requests for training purposes of any platform
View as text
Compare with Figure 3-7 (OpenAI); read the requests arriving at the site and the platform's internal live-test topics separately.
Anthropic: 24,479,305 requests for training purposes Finding: Almost all reads are for training purposes. It has the highest number of requests for training purposes of any platform. The Anthropic (Claude) path: from crawlers to platform to customers 2026/3/3–10/8 · five crawlers, 24,492,552 requests total S1 Prepare S2 Crawl S3 Train S4 Index S5 Ask AI S6 AI Mention S7 AI Answer S8 Act S9 Visit S10 Return Site-arrival requests by four purposes Training, indexing, and real-time retrieval, plus 48 requests with purpose still being tracked; each item is counted once. S3 Train 24,479,305 requests S4 Index 10,705 requests S5 Ask 2,494 requests Black Box: inside AI Platforms S6–S8 Mentions, answers, and actions Tracking in Progress S1 Prepare Our website S2 Crawl Anthropic's five crawlers S3–S5 Site-arrival purposes Site request log S6–S8 Inside the platform Black Box S9 Visit Organic Traffic Whole site 196,702,526 Requests · 2/12–10/8 Anthropic 24,492,552 Requests S3 Train Training crawler 24,479,305 Requests Takes content back for model training S4 Index Search crawler 10,705 S5 Ask Real-time retrieval crawler 2,494 Site-arrival requests visible In AI Platforms: live tests Indexing requests visible In AI Platforms: live tests Real-time retrievals visible In AI Platforms: live tests After training In AI Platforms: live tests S6 Mention In AI Platforms: live tests S7 Answer In AI Platforms: live tests S9 Visit Organic Traffic (own) Tracking in Progress Site-wide total 309,826 visits S8 Act Tracking in Progress S10 Return Tracking in Progress On this path, what can be counted directly is crawler requests arriving at the site. S3 24,479,305 + S4 10,705 + S5 2,494 + purpose being tracked 48 = 24,492,552. Crawler requests and the platform's mentions, answers, and actions are each read in their own units. In-platform processing in S6–S8 and the S9 Organic Traffic breakdown are still being tracked. Sources: site-wide and Organic Traffic from approved figures; each platform's crawler requests from the 2026/3/3–10/8 approved list. Green bands show the purposes of site-arrival requests; dashed boxes mark in-platform or platform-specific numbers still being tracked. ID EO
How to readCompare with Figure 3-7 (OpenAI); read the requests arriving at the site and the platform's internal live-test topics separately.
Figure 3-9 Apple: 24,615,429 indexing requests｜Highest indexing volume of any platform
View as text
Look at crawl volume first, then follow the journey to see the tracking scope of later use.
Apple: indexing requests reach 24,615,429 Finding: all reads go to building the search index (Siri, Spotlight, and other search features); the highest indexing volume of any platform. The Apple path: from crawlers to platform to customers 2026/3/3–10/8 · one crawler, 24,615,429 requests total S1 Prepare S2 Crawl S3 Train S4 Index S5 Ask AI S6 AI Mention S7 AI Answer S8 Act S9 Visit S10 Return Site-arrival requests for indexing All requests with an assigned purpose are in S4 Index; this batch is counted once. S3 Train Tracking in Progress S4 Index 24,615,429 requests S5 Ask Tracking in Progress Black Box: inside AI Platforms S6–S8 Mentions, answers, and actions Tracking in Progress S1 Prepare Our website S2 Crawl Apple's single crawler S3–S5 Site-arrival purposes Site request log S6–S8 Inside the platform Black Box S9 Visit Organic Traffic Whole site 196,702,526 Requests · 2/12–10/8 Apple 24,615,429 Requests S3 Train Platform-specific requests Tracking in Progress Takes content back for model training S4 Index Search crawler 24,615,429 S5 Ask Platform-specific requests Tracking in Progress Tracking in Progress In AI Platforms: live tests Indexing requests visible In AI Platforms: live tests Tracking in Progress In AI Platforms: live tests After training In AI Platforms: live tests S6 Mention In AI Platforms: live tests S7 Answer In AI Platforms: live tests S9 Visit Organic Traffic (own) Tracking in Progress Site-wide total 309,826 visits S8 Act Tracking in Progress S10 Return Tracking in Progress On this path, what can be counted directly is crawler requests arriving at the site. The 24,615,429 requests in S4 equal the Apple crawler total. Crawler requests and the platform's mentions, answers, and actions are each read in their own units. Apple's platform-specific numbers for S3 and S5–S10 are still being tracked. Sources: site-wide and Organic Traffic from approved figures; each platform's crawler requests from the 2026/3/3–10/8 approved list. Green bands show the purposes of site-arrival requests; dashed boxes mark in-platform or platform-specific numbers still being tracked. ID EO
How to readLook at crawl volume first, then follow the journey to see the tracking scope of later use.
Figure 3-10 Meta: 11,583,411 requests for training purposes｜All reads are for training purposes
View as text
Look at AI crawls and customer arrivals separately, reading each result in its own unit.
Meta: 11,583,411 requests for training purposes Finding: All reads are for training purposes. Separately, Organic Traffic totaled 9,388 visits over the same period. Meta's path: from crawler to platform, then to customers 2026/3/3–10/8 · one crawler, 11,583,411 requests in total S1 Prepare S2 Crawl S3 Train S4 Index S5 Ask AI S6 AI Mention S7 AI Answer S8 Act S9 Visit S10 Return Site requests for training use All requests with an assigned purpose fall under S3 Train; this batch is counted only once. S3 Train 11,583,411 requests S4 Index Tracking in Progress S5 Ask Tracking in Progress Black Box: inside AI Platforms S6–S8 Mentions, answers, and actions Tracking in Progress S1 Prepare Our website S2 Crawl One Meta crawler S3–S5 Site-arrival purposes Site request log S6–S8 Inside the platform Black Box S9 Visit Organic Traffic Whole site 196,702,526 Requests · 2/12–10/8 Meta 11,583,411 Requests S3 Train Training crawler 11,583,411 Requests Takes content back for model training S4 Index Platform-specific requests Tracking in Progress S5 Ask Platform-specific requests Tracking in Progress Site-arrival requests visible In AI Platforms: live tests Tracking in Progress In AI Platforms: live tests Tracking in Progress In AI Platforms: live tests After training In AI Platforms: live tests S6 Mention In AI Platforms: live tests S7 Answer In AI Platforms: live tests S9 Visit Organic Traffic 9,388 visits Site-wide total 309,826 visits S8 Act Tracking in Progress S10 Return Tracking in Progress On this path, what can be counted directly is crawler requests arriving at the site. The 11,583,411 requests at S3 equal the Meta crawler total. Crawler requests and the platform's mentions, answers, and actions are each read in their own units. S9 Organic Traffic is a separate 9,388 visits; read requests and visits by humans each in their own unit. Sources: site-wide and Organic Traffic from approved figures; each platform's crawler requests from the 2026/3/3–10/8 approved list. Green bands show the purposes of site-arrival requests; dashed boxes mark in-platform or platform-specific numbers still being tracked. ID EO
How to readLook at AI crawls and customer arrivals separately, reading each result in its own unit.
Key Points of This Chapter
| Claim | Why |
|---|---|
| The AI black box is the training, ranking, mentions, answers, and actions that happen inside AI platforms. | Mentions, answers, and actions in S6–S8 happen inside the platform. The Black Box also hides whether content read in S3–S4 is used for training or indexed, and how it is ranked. Being read does not mean content has been used for training or indexed. The website records crawls and site arrivals directly. S5 real-time retrieval also has first-party numbers. In the figure, the S6 number is search impressions (not AI mentions). The mention rate in AI answers will be published later. |
| IDAEO measures both ends of the black box, then uses platform data and periodic live tests to measure the results of content use. | Crawls before entering and site arrivals after leaving are both first-party measurements. In the figure, S6 lists search impressions (not AI mentions), which come from platform data. The IDAEO platform's 31,700 Copilot citations also come from platform data. The mention rate in AI answers will be published later. Periodically asking multiple AIs directly can confirm whether you were mentioned and whether the answers were correct. Each source provides different evidence, but none reveals the platform's internal processing. |
| Periodic live tests of TK Lin's personal IDA record whether it appears in AI search source lists. | Live test of TK Lin's personal IDA: on Sep 20, launch day, AI citations were 0. On Oct 3, a live test covered 100 questions across five AI models. This IDA appeared in AI search source lists at a rate of 84% (four of the five models used the same search interface; the questions named the author himself; a single case). It ranked first among all sources by number of appearances. The 0 on Sep 20 is a citation count; the 84% on Oct 3 is a source-list appearance rate. They use different units and cannot be compared directly. This is a single case and cannot be generalized to merchants. Platform-wide numbers for AI Mention Rate and AI Answer Accuracy Rate will be published later. |
| Results from each source are read separately by unit. | Requests, pages, and clicks represent different events; only comparisons on the same basis are meaningful for marketing. |
| The platform AI Citations figure published in this book comes from Copilot: the IDAEO platform has been cited 31,700 times since Jul 22. | This is the citation result provided by Copilot; the remaining platforms are still being tracked and will be published later. Label the source and period clearly so comparisons are made on the same basis. |
Terms
- Black Box
- Mentions, answers, and actions in S6–S8 happen inside AI Platforms. The Black Box also hides whether content read in S3–S4 is used for training or indexed, and how it is ranked. Being read does not mean content has been used for training or indexed. IDAEO observes results through measurements at both ends and periodic live tests. In the figure, the S6 number is search impressions (not AI mentions). The mention rate in AI answers will be published later.
- Platform Data
- Platforms such as Google and Bing provide indexing, search impressions, and clicks. S6 in the figure is search impressions (not AI mentions). The platform AI citation data in this book comes from Copilot. The mention rate in AI answers will be published later.
- Indexed
- The result of a search platform adding pages to its index, counted in pages.
- Search result impressions
- The number of times a page appears in search results. The S6 cell in the figure lists search impressions (not AI mentions). The mention rate in AI answers will be published later.
FAQ
Q: How do you know AI cites you?
A: Look at three kinds of evidence together. First, citation counts reported by platforms. The figures published in this book record Copilot citations of the IDAEO platform. Since Jul 22, the total is 31,700 citations. Second, regularly ask multiple AI services questions to check mentions and answer accuracy. Third, records of source lists for individual cases. TK Lin's personal IDA had 0 AI citations on its launch day, Sep 20. On Oct 3, testing covered 100 questions and five AI models. This IDA appeared in AI search source lists at a rate of 84%. The 0 is a citation count; 84% is the share of source lists in which it appeared. The units differ, so they cannot be compared directly. This is a single case and cannot be generalized to merchants. There is another signal: first-party records show AI real-time reads totaling 651,000. These are reads, not necessarily citations. Citations occur inside AI Platforms, and these records provide different kinds of evidence. See Figure 3-6, “How Do We Know AI Cited Us?”, for an overview. Also see Figure 3-2, “Content Journey Map and the AI Black Box,” and Figure 3-5, “Microsoft.”
Q: Where do Google's and OpenAI's numbers come from?
A: Google's indexing, search impressions, and clicks come from the search platform. S6 in the figure is search impressions (not AI mentions). The mention rate in AI answers will be published later. OpenAI's retrievals and customer arrivals come from first-party measurement. The data scopes differ, so check source and step first when placing them side by side.
Q: Can IDAEO open the AI black box?
A: We cannot open it, but we can measure both ends. Mentions, answers, actions, and the training and ranking that follow crawls happen inside AI Platforms. Measure crawls and site arrivals at both ends, alongside search impressions, Copilot citations, and periodic questions. In the figure, S6 lists search impressions (not AI mentions). These data can provide evidence of citations, mentions, and answer accuracy. Live test of TK Lin's personal IDA: on Sep 20, launch day, AI citations were 0. On Oct 3, a live test covered 100 questions across five AI models. This IDA appeared in AI search source lists at a rate of 84%. It ranked first among all sources by number of appearances. The 0 is a citation count; the 84% is a source-list appearance rate. They use different units and cannot be compared directly. This is a single case and cannot be generalized to merchants. This evidence still does not reveal the platform's internal processing. Platform-wide numbers for AI Mention Rate and AI Answer Accuracy Rate will be published later.
Q: Who is counted in this chapter's numbers?
A: Platform results are presented separately by source and what was measured. The Copilot citation count covers the IDAEO platform. TK Lin's personal IDA is a live test of a single piece of content; the percentage refers to its appearance in AI search source lists. The subjects and measurement methods differ. Each result must be interpreted within its published scope.
Q: Bing has indexed 375,000 pages and Google 282,000. How should you compare them?
A: Look at the two separately; each represents that platform's indexing result. The same page gets indexed by both, so adding them together double-counts.
This Chapter in Three Sentences
- Mentions, answers, and rankings inside AI Platforms are hidden in a Black Box.
- We can measure both ends: reads going in and site arrivals coming out, supplemented by periodic questions to AI.
- Reads, search impressions, and citations are different events and cannot be added together.
Three Things You Can Do This Week
- Organize your website's read and visit records, clearly labeling each source and period.
- Review the indexing, search impression, and citation records provided by search platforms. Record them separately.
- Ask AI about your business, then check its answers and listed sources.
Next ChapterIn the next chapter, you will learn which three types of sources are used to verify merchant information.
Self-CheckIdentify the Black Box: S6–S8, and what happens after the S3–S4 crawls. Explain what evidence first-party measurement at both ends, platform data, and periodic live tests each provide. Then use TK Lin's personal IDA to distinguish source-list records from an individual case from platform metrics. When reading S6 in the figure, make clear that the number is search impressions (not AI mentions). The mention rate in AI answers will be published later.
Free to download, updated monthly
Cite this book
IDAEO: Marketing in the Age of AI · 林毅祥（TK Lin） · 1st edition, 9th revision · · https://km.idaeo.ai/idaeo/textbook?lang=en · CC BY 4.0