FR
Documentation

Where the figures come from and why they differ

GA4, Search Console, Bing, Kaiku scan, external check: what each source measures, and why two correct figures can differ

Open the page in the app
On this page

The rule: real data is the reference, the scan explains why

The same assistant can show three different figures in Kaiku: the visits it sends you (GA4), the share of Kaiku scan answers that name you, and the DataForSEO external check. All three are right: they do not measure the same thing, nor on the same questions. Real data (GA4, Search Console, Bing, Google listing) says what really reaches your site, whatever the question: that is what counts. The Kaiku scan asks AI engines YOUR tracked prompts: it explains why you are cited or not on those questions, and what to change.

  • Real (what really reaches your site — the reference) : GA4 · real · Search Console · real · Bing · real · Business Profile · real.
  • Diagnosis (explains why, on a sample you choose) : Kaiku scan · your prompts · AI Overview · your keywords · AI Overview · your prompts · AI Overview · your queries · AI market mentions.
  • Check (a second look, independent of the scan) : DataForSEO · external check.

GA4 · real — Google Analytics 4 — real visits

The visits AI assistants (ChatGPT, Perplexity, Gemini, Claude, Copilot…) really send to your site, whatever the question.

  • On what: Every GA4 session whose source (sessionSource) is an AI assistant, over the snapshot window.
  • How often: Read at the Sunday scan and when Rank › GA4 is opened (at most every 6 hours).
  • Reliability: Real data: this is the reference.

Search Console · real — Google Search Console — real data

Google impressions shown in an AI answer (imported “generative AI” report) and your Google clicks.

  • On what: Every Google query that shows your site, as Google counts it.
  • How often: At each import of the generative AI report (Google offers no API for it); clicks read daily, with Google's publication delay.
  • Reliability: Real data: this is the reference.

Bing · real — Bing Webmaster — real Copilot citations

Your site's citations in Copilot answers (Bing “AI Performance” export).

  • On what: Every Copilot answer citing your pages, as Bing counts them.
  • How often: At each import of the AI Performance export (about 6 months of history at Bing).
  • Reliability: Real data: this is the reference.

Business Profile · real — Google Business Profile — real data

Views, searches and actions (calls, directions, website visits) on your Google listing.

  • On what: Your verified listing, as Google measures it.
  • How often: Read through the Google Business Profile API, a few days behind Google.
  • Reliability: Real data: this is the reference.

Kaiku scan · your prompts — Kaiku scan — diagnosis on your tracked prompts

What AI engines answer to YOUR tracked prompts: the share of usable answers that name your brand.

  • On what: Your tracked prompts, asked in the project's language; 1 query per prompt and engine (2 at the Sunday scan without web search).
  • How often: Once a week (Sunday automatic scan), plus scans started by hand. On Sundays, ChatGPT, Gemini and Claude search the web before answering on the Pro and Enterprise plans; scans started by hand stay without web search.
  • Reliability: Diagnosis: explains why, does not replace real data.

DataForSEO · external check — DataForSEO external check

ChatGPT, Perplexity and Gemini answers to your SEO keywords asked as they are, collected by DataForSEO with its own models and web search: is the brand named?

  • On what: A sample of at most 10 keywords, one answer per keyword.
  • How often: Read after the Sunday scan and on “Refresh” (billed per keyword).
  • Reliability: Check: a second look, independent of the Kaiku scan.

AI Overview · your keywords — AI Overview — your keywords (DataForSEO)

Whether Google's AI summary shows on your SEO keywords, and whether it cites your site as a source.

  • On what: Your tracked SEO keywords, read by DataForSEO on Google in the project's country and language.
  • How often: Read after the Sunday scan and on “Refresh”.
  • Reliability: Diagnosis: explains why, does not replace real data.

AI Overview · your prompts — AI Overview — your prompts (Kaiku scan)

Google's AI summary queried with your tracked prompts: does it name you?

  • On what: Your tracked prompts, asked to Google in the project's language.
  • How often: Once a week (Sunday automatic scan), plus scans started by hand. On Sundays, ChatGPT, Gemini and Claude search the web before answering on the Pro and Enterprise plans; scans started by hand stay without web search.
  • Reliability: Diagnosis: explains why, does not replace real data.

AI Overview · your queries — AI Overview — your Search Console queries

On your Search Console queries that get clicks, does Google's AI summary cite your site or a competitor?

  • On what: Search Console queries with clicks, checked by DataForSEO.
  • How often: At each “clicks vs AI answers” reading.
  • Reliability: Diagnosis: explains why, does not replace real data.

AI market mentions — AI market mentions (DataForSEO LLM Mentions)

How many Google AI answers (and ChatGPT answers for the United States) mention your domain and your competitors' on your market.

  • On what: DataForSEO's base of AI answers for your market, well beyond your tracked keywords.
  • How often: Every Sunday.
  • Reliability: Diagnosis: explains why, does not replace real data.

Example 1 — Visits from ChatGPT, but ChatGPT at 0% in the scan

GA4 counts about a hundred visits from chatgpt.com in 90 days, mostly to the home page and “About”. The Kaiku scan shows ChatGPT at 0% on 5 generic prompts (“best kitchen renovator in Montreal”…). Both are right: these visitors asked ChatGPT about your business by name, and ChatGPT gave them your site; your tracked prompts ask category questions where it names other brands. A visit to a blog article, by contrast, comes from an information question where ChatGPT cites your article.

  • Where to see it: click the ChatGPT row in “Visibility by AI engine” or in Score by engine: the engine's sheet puts its GA4 visits, its landing pages, the scan prompt by prompt and the external check side by side, with a sentence explaining the gap from your data.
  • What to do: add prompts close to these visitors' questions (the landing pages show the topics); Kaiku never creates any on your behalf.
  • Web search: a scan without web search queries ChatGPT “from memory”, whereas the consumer ChatGPT searches the web. On the Pro and Enterprise plans, the Sunday scan turns web search on for ChatGPT, Gemini and Claude.

Example 2 — The scan names you, the external check is at 0

The Kaiku scan shows Perplexity at 40% on your tracked prompts, but the DataForSEO external check names you on none of its 10 keywords. The questions differ: the scan asks your prompts, phrased as real questions, in the project's language; the external check asks your SEO keywords as they are (“kitchen renovation”), with DataForSEO's models, at another time. A two-word keyword does not trigger the same answer as a full question.

  • The criterion is the same on both sides: the brand NAMED in the answer. “Cited as a source” is extra information.
  • The external check is a second look: a lasting gap says your content answers long questions but not short searches (or the reverse).

Example 3 — AI Overview: your prompts or your keywords

Kaiku reads Google's AI summary in two ways. “AI Overview — your prompts” is queried by the scan with your tracked prompts: it says whether Google names you when asked those questions. “AI Overview — your keywords” is read by DataForSEO on your SEO keywords: it says whether the summary shows on those searches, and whether it cites your site as a source. The two can diverge: Google does not show a summary on every short search, and a long question does not have the same sources as a keyword.

  • The real data remains Search Console's “generative AI” report: your site's impressions in Google's AI answers, across all queries.