How every number on this site is measured, and what it does not mean
Each figure comes from an export with a date, is processed by a script that records its file, sheet and calculation, and is published as text, in a chart drawn from the same rows, and in a table. This page is the contract behind that.
The chain, one step at a time
A visitor decides in this order, and each step is a different measurement from a different tool. Nothing is converted into the next step.
Definitions, by platform
Google Search Console, web search. Impression: the site appeared in a Google results page for a query. Click: the user clicked it. CTR: clicks divided by impressions. Countries: the searcher's country as Google records it. The exports say "last 16 months" but each property only holds data from the day it was created, so the zero days before the first impression are absence of data, not measured absence of demand.
Google Search Console, generative AI features. The number of times the site appeared inside Google's AI experiences (AI Overviews and AI Mode). It is an impression count with no clicks attached. It is written here as "impressions inside Google AI features", never as recommendations, visits or leads.
Bing Webmaster Tools, search performance. Impressions and clicks in Bing. The country report in the export lists ten fixed countries plus an unspecified bucket, so it is shown as is and never added to the country unions.
Bing Webmaster Tools, AI performance. Citations: how many times pages of the site were used as a source in Microsoft AI answers. Cited pages: a per-day count that cannot be summed into unique pages, so it is not summed. A citation is a reference, not a click and not an endorsement.
Google Analytics 4. Sessions and users as GA4 counts them. Sessions from AI assistants: sessions whose recorded source is chatgpt.com, copilot.com, perplexity, gemini.google.com or another AI platform, or whose medium is ai-assistant. This proves the visit's origin, not what the assistant said. Key events: actions each site marks as conversions; they differ between sites and are not compared across sites. Users in data-centre cities (Ashburn, Council Bluffs, Boardman) are likely automated and are not claimed as audience.
How the totals are built
- Only figures of the same nature are added: Google impressions with Google impressions, citations with citations. Never impressions plus clicks plus sessions.
- Countries are a set union: a country reached by three sites counts once. Google (web plus AI) and GA4 unions are kept separate. Portfolio figures: 133 countries in Google, 53 in GA4.
- Windows differ by property because each property was created on a different date. Every figure carries its window. The shared window (23 Jul to 19 Sep 2026, 59 days) is published on the proof page next to the full-window totals.
- Before/after compares the first 28 days of a property with the last 28 days ending 19 Sep 2026: same length, no overlap. Growth percent is (after minus before) divided by before, times 100. The multiplier is after divided by before. When before is zero, no percentage is computed and the text says "not detected in the first 28 days".
- Where a property was created with the site (the usual case), the baseline is the property's first weeks, which is stated. Sanick's Bing history is the one true pre-intervention baseline and is compared year over year.
- Attribution language: growth observed after the structure went live, in the period tracked. Causality is asserted only where the intervention is documented and no competing factor (ads, domain change, campaign, seasonality) is known. Amo Embarque's domain change is documented on its page.
- Estimates are never printed as measured numbers. If something is an estimate, it is not on the site.
Provenance and reproducibility
The raw exports live in a private folder that is excluded from the public repository. A script reads them and writes two files: the public data set and a provenance log with one entry per published figure (company, file, sheet, rows, calculation, processing time). The site pages, charts and tables are generated from the public data set, so a number cannot differ between the text, the chart and the table. The public data set is at /proof/portfolio-data.json; it contains aggregates only: no names, contacts, revenue or individual-level records.
Logos, flags and platform names
Client logos are the companies' own marks, taken from their own sites or brand files, unaltered, shown to identify the client. Flags are the ISO 3166 set from the open-source flag-icons project (MIT). Country names are always present as text; flags are decoration. Platform names (Google Search Console, Bing Webmaster Tools, Google Analytics, ChatGPT, Copilot) identify the source of a measurement and imply no partnership, certification or endorsement.
Update cycle
Exports are collected, processed and published together. This publication uses exports of 21 and 22 Sep 2026: the Search Console, Bing and most GA4 files were downloaded on 21 Sep 2026; GA4 source and city reports for three properties and one previous property, and two Bing AI reports, were added on the next day. Every daily series ends on 19 Sep 2026; GA4 totals run to the export date shown on each case. Previous publications used exports of 17 Sep 2026 and were replaced in full.
Same standard for your numbers.
If you want your growth measured this way, with the export date on every figure, start with the visibility check.
Run my visibility check