A voice AI language product I built and run, published from the starting line: 42 countries in 80 days and 90 key events
The method applied where I am the client. FalaMandra is a voice-first language product: the learner talks to an AI tutor, is corrected in the flow, and buys minutes when the free hour runs out. The numbers are small and the page says so. It is here because a portfolio that only shows finished curves hides the part that matters: where a curve starts and how it is measured.

- 725impressions in Google Search, 13 clicksSearch Console · 2 Jul to 19 Sep 2026
- 29impressions inside Google generative AI featuresSearch Console AI report · same window · 8 countries
- 42countries where the site appeared in Google resultsSearch Console · 3 of them clicked
- 88citations as a source in Microsoft AI answersBing Webmaster Tools · 13 Jul to 19 Sep 2026
- 21sessions from AI assistants, 3% of 610GA4 · 1 Jan to 21 Sep 2026
- 20countries with users in GA4, 90 key eventsGA4 · since launch
01 · ContextThe business
Brazilians learning English and French who understand the language and freeze when they have to speak. Own product: method, agent, voice pipeline, checkout and site, all built and operated by me.
02 · Starting pointWhere it started
Search Console property created on 2 Jul 2026. First 28 days: 47 Google impressions. No ads at any point.
03 · ChallengeThe problem to solve
Get a brand-new product with no brand to appear for the questions learners type, in two languages, and measure whether sessions from AI assistants behave differently from the rest.
04 · StrategyWhat the plan was
Two content systems, English and French, written on real learner questions, plus GA4 key events on the actions that matter (account created, minutes bought), so the origin of a session can be compared with what it did.
05 · StructureWhat I built
- The teaching method, the conversational agent, the voice pipeline and the correction logic.
- The product: accounts, dashboard, minute packages, checkout and confirmation, on a static site with serverless functions.
- Two content systems, English and French, written on the questions learners actually type.
- GA4 key events on the actions that matter.
06 · TimelineDated milestones
- 2 Jul 2026Search Console property created.
- 5 Jul 2026First Google impression.
- 25 Jul 2026First impression inside Google generative AI features.
- 8 Sep 2026Peak day in Google Search so far: 32 impressions.
- 19 Sep 2026Last complete day in the 21 Sep 2026 export.
07 · From impression to actionEach step, its own measurement
Impressions are appearances in a search or AI experience. Clicks are recorded at the source. Sessions are recorded on the site. Key events are actions the site tracks. Leads and deals live in the client's CRM. None of these is converted into another.
08 · Before and afterSame-length windows, no overlap
949% growth
Google Search impressions, Search Console. Baseline is the first 28 days after the property was created; no earlier search data exists for this property. Same-length windows, no overlap, 80 days from the start of the first to the end of the second. Growth observed after the structure was published.
2,400% growth
Impressions inside Google generative AI features, Search Console AI report, same windows. Same-length windows, no overlap, 80 days from the start of the first to the end of the second.
3,000% growth
Bing Webmaster Tools, search performance, same windows. Same-length windows, no overlap, 70 days from the start of the first to the end of the second.
09 · EvolutionDrawn from the daily export, nothing smoothed
Google Search impressions per day
Search Console, web search, 2 Jul 2026 to 19 Sep 2026 (80 days). The dashed line marks search console property created.
Show the numbers as a table
| Month | Days in export | Impressions | Per day |
|---|---|---|---|
| Jul 2026 (partial) | 30 (2 Jul to 31 Jul) | 56 | 1.9 |
| Aug 2026 | 31 (1 Aug to 31 Aug) | 291 | 9.4 |
| Sep 2026 (partial) | 19 (1 Sep to 19 Sep) | 378 | 19.9 |
Google Search impressions per month
Same export, summed by calendar month. Partial months are shown as exported and marked.
Show the numbers as a table
| Month | Days | Impressions |
|---|---|---|
| Jul 2026 (partial) | 30 | 56 |
| Aug 2026 | 31 | 291 |
| Sep 2026 (partial) | 19 | 378 |
Impressions inside Google generative AI features per day
Search Console, generative AI features report, 2 Jul 2026 to 19 Sep 2026 (80 days). First impression on 25 Jul 2026.
Show the numbers as a table
| Month | Days in export | Impressions | Per day |
|---|---|---|---|
| Jul 2026 (partial) | 30 (2 Jul to 31 Jul) | 2 | 0.1 |
| Aug 2026 | 31 (1 Aug to 31 Aug) | 2 | 0.1 |
| Sep 2026 (partial) | 19 (1 Sep to 19 Sep) | 25 | 1.3 |
Citations as a source in Microsoft AI answers, per week
Bing Webmaster Tools, AI performance report, 13 Jul 2026 to 19 Sep 2026 (69 days), summed in 7-day buckets.
Show the numbers as a table
| Month | Days in export | Citations | Per day |
|---|---|---|---|
| Jul 2026 (partial) | 19 (13 Jul to 31 Jul) | 0 | 0.0 |
| Aug 2026 | 31 (1 Aug to 31 Aug) | 39 | 1.3 |
| Sep 2026 (partial) | 19 (1 Sep to 19 Sep) | 49 | 2.6 |
10 · GeographyWhere the demand appeared
42 countries with at least one impression in Google Search; 3 of them produced a click. Impressions mean the site appeared; clicks mean someone chose it. The two are shown separately.
All countries as a table (ISO code, sorted by impressions)
| # | ISO | Country | Impressions | Clicks |
|---|---|---|---|---|
| 1 | BR | Brazil | 425 | 11 |
| 2 | US | United States | 50 | 0 |
| 3 | PT | Portugal | 35 | 0 |
| 4 | FR | France | 32 | 0 |
| 5 | AO | Angola | 26 | 1 |
| 6 | CA | Canada | 18 | 0 |
| 7 | ES | Spain | 14 | 0 |
| 8 | GB | United Kingdom | 13 | 0 |
| 9 | BE | Belgium | 12 | 0 |
| 10 | CV | Cape Verde | 11 | 0 |
| 11 | MZ | Mozambique | 10 | 1 |
| 12 | DE | Germany | 9 | 0 |
| 13 | IT | Italy | 8 | 0 |
| 14 | CH | Switzerland | 8 | 0 |
| 15 | MX | Mexico | 7 | 0 |
| 16 | IN | India | 6 | 0 |
| 17 | NL | Netherlands | 5 | 0 |
| 18 | CO | Colombia | 3 | 0 |
| 19 | LU | Luxembourg | 3 | 0 |
| 20 | PH | Philippines | 3 | 0 |
| 21 | AR | Argentina | 2 | 0 |
| 22 | AU | Australia | 2 | 0 |
| 23 | DK | Denmark | 2 | 0 |
| 24 | IE | Ireland | 2 | 0 |
| 25 | SE | Sweden | 2 | 0 |
| 26 | BA | Bosnia & Herzegovina | 1 | 0 |
| 27 | CL | Chile | 1 | 0 |
| 28 | CD | Congo - Kinshasa | 1 | 0 |
| 29 | DO | Dominican Republic | 1 | 0 |
| 30 | EE | Estonia | 1 | 0 |
| 31 | GR | Greece | 1 | 0 |
| 32 | ID | Indonesia | 1 | 0 |
| 33 | JP | Japan | 1 | 0 |
| 34 | MY | Malaysia | 1 | 0 |
| 35 | NP | Nepal | 1 | 0 |
| 36 | NG | Nigeria | 1 | 0 |
| 37 | NO | Norway | 1 | 0 |
| 38 | PE | Peru | 1 | 0 |
| 39 | BL | St. Barthélemy | 1 | 0 |
| 40 | TH | Thailand | 1 | 0 |
| 41 | UZ | Uzbekistan | 1 | 0 |
| 42 | VN | Vietnam | 1 | 0 |
Google Search Console, sheet Países, same window as above. ISO 3166 codes in the table.
Countries that actually visited (GA4)
20 countries with at least one active user, 2 with a key event. A GA4 user is a device or browser as the tool counts it, not a verified person.
All countries as a table (ISO code, sorted by users)
| # | ISO | Country | Users | Key events |
|---|---|---|---|---|
| 1 | US | United States | 123 | 0 |
| 2 | BR | Brazil | 79 | 79 |
| 3 | CN | China | 13 | 0 |
| 4 | DE | Germany | 6 | 0 |
| 5 | CA | Canada | 5 | 0 |
| 6 | NL | Netherlands | 5 | 0 |
| 7 | FR | France | 3 | 0 |
| 8 | NZ | New Zealand | 3 | 0 |
| 9 | PT | Portugal | 3 | 0 |
| 10 | GB | United Kingdom | 3 | 0 |
| 11 | IE | Ireland | 2 | 0 |
| 12 | KR | South Korea | 2 | 0 |
| 13 | AO | Angola | 1 | 0 |
| 14 | EE | Estonia | 1 | 0 |
| 15 | IN | India | 1 | 0 |
| 16 | MZ | Mozambique | 1 | 0 |
| 17 | PL | Poland | 1 | 0 |
| 18 | CH | Switzerland | 1 | 11 |
| 19 | TW | Taiwan | 1 | 0 |
| 20 | TH | Thailand | 1 | 0 |
Google Analytics 4, Demographic details: Country, 1 Jan 2026 to 21 Sep 2026. 8 users had no country set. Users in data-centre cities (Ashburn, Council Bluffs, Boardman) are likely automated and are not claimed as audience.
Bing, top countries
Bing Webmaster Tools reports a fixed list of ten countries plus an unspecified bucket; it is not a full country list and is not added to the counts above. Brazil 69, United States 18, United Kingdom 4, France 4, China 1, Canada 1; unspecified 16.
11 · AI visibilityThree different measurements, kept apart
Impressions inside Google generative AI features: 29 in 80 days, from 8 countries, first one on 25 Jul 2026, peak 4 on 14 Sep 2026. This is Search Console's own report of the site appearing inside AI Overviews and AI Mode. It counts appearances, not clicks or visits.
Pages surfaced most:
/blog/quanto-custa-aprender-frances/9 AI impressions/7 AI impressions/blog/conversacao-em-frances-para-iniciantes/3 AI impressions/ingles/3 AI impressions/blog/curso-de-frances-online-gratis/2 AI impressions
Citations as a source in Microsoft AI answers: 88 between 13 Jul and 19 Sep 2026, first on 19 Aug 2026, up to 3 different pages cited in a single day. From Bing Webmaster Tools' AI performance report. A citation is the page being used as a reference in a generated answer; it is not a click.
Traffic detected from AI platforms: 21 sessions whose recorded source is an AI assistant, chatgpt.com / ai-assistant 15, copilot.com / ai-assistant 5, copilot.com / (not set) 1. From GA4 session source/medium. This proves the origin of the visit, not the content of the conversation.
12 · TrafficWhat arrived on the site
610 sessions since launch, grouped by source family. Referral includes the previous domain where one exists. Every row is text, so the split is readable without the bars.
GA4, session source / medium, 1 Jan 2026 to 21 Sep 2026. Active users 263, new users 267, average engagement 159.2 s per active user.
Top cities by users: Chapeco 30, Sao Paulo 17, Porto Alegre 12, Glenview 10, Belo Horizonte 6, New York 6.
13 · ReadingWhat changed
Google impressions went from 47 in the first 28 days to 493 in the last 28, across 42 countries. France produced 32 impressions for French-learning queries. GA4 recorded 21 sessions from AI assistants (chatgpt.com 15, copilot.com 6) and 90 key events, all of them from Brazil.
14 · LimitsThe honest part
13 clicks in 80 days is a signal, not traffic. Most sessions are direct, which is the founder's own network and testing. GA4 lists 123 users from the United States, but their engagement rate is near zero and the cities are data-centre locations (Council Bluffs, Boardman, Ashburn); that is automated traffic and is not claimed as an audience. Google's AI features showed the site only 29 times. The point of publishing now is the starting line, with dates, so the curve can be checked as it forms.
15 · Method and sourcesHow to read these numbers
Attribution
Growth is reported as observed after the structure went live, in the period tracked. Where the property was created with the site, the baseline is the first 28 days of the property, which is stated on every comparison. No paid traffic was run on any of these sites in the window. Sales, leads and revenue belong to the client and are not published. Full method: how the numbers are measured. Machine-readable data: portfolio-data.json.
- Google Search ConsoleWeb search performance, 2 Jul 2026 to 19 Sep 2026 (80 days). Impressions, clicks, countries, pages, queries, devices.
- Google Search ConsoleGenerative AI features report, 2 Jul 2026 to 19 Sep 2026 (80 days). Impressions only; the report has no clicks.
- Bing Webmaster ToolsSearch performance, 12 Jul 2026 to 19 Sep 2026 (70 days). Impressions and clicks; country report limited to ten fixed countries.
- Bing Webmaster ToolsAI performance report, 13 Jul 2026 to 19 Sep 2026 (69 days). Citations of the site's pages as a source in Microsoft AI answers.
- Google Analytics 4Property FalaMandra, 1 Jan 2026 to 21 Sep 2026 (export window; data exists from launch; last day partial). Users, sessions by source, key events, countries, cities.
- CollectionFiles downloaded on 21 and 22 Sep 2026. Search Console exports end on 19 Sep 2026 because the report lags. Processed by a script that keeps a provenance entry per figure.
Logos are the companies' own marks, shown to identify the client. Platform names identify the source of the data and imply no partnership or endorsement.
Questions people with this problem ask
Why publish a case with such small numbers?
Because the method is the claim, not the total. Every client case on this site started at zero on a dated export. Showing my own product at that stage is the honest version of the portfolio.
Does the product get traffic from AI assistants?
GA4 recorded 21 sessions from AI assistants between 1 Jan 2026 and 21 Sep 2026, 3% of all sessions. They are visits with a recorded origin; nothing more is inferred.
I test the method on my own money first.
If you want the same measurement discipline on your business, with the numbers published at the stage they are actually at, send me the site.
Run my visibility check