AIDOL
A public experiment

Experiment diary

For 90 days I'm building an AI model to her first earned ruble and logging every step: numbers, prompts, failures. Every figure is real, including zero.

Where the experiment stands now

updated: October 5
23 / 90
Day
84
model's followers on Threads
3
model's followers on Instagram
20
author's followers on Threads
≈17 900 ₽
spent since Sep 17
0 ₽
earned in total
Trajectory: followers and spending by day
model's followers on Threadsspent cumulatively, ₽
0D511020D6141041D7161041D8193454D9203360D16505071D181642683D221729184D2317879

Income has been zero throughout, so it gets no line of its own. Spending is counted from September 17, when daily tracking began.

Day 23 / 90October 5, 2026

Day 23. The site moved to a Russian server and opens without a VPN, while the model's Instagram posts didn't go out for five days because of empty rows in the spreadsheet

84 / 3
model's followers: Threads / Instagram (was 83 / 3)
≈590 ₽
spent today on frame generations; ≈17,900 ₽ since Sep 17
5
author's hours today
0 ₽
earned

On Day 22 I promised three things. I did not take the follows for the two test frames on October 5; the test is not closed. The two-week check of the new rules starts tomorrow, October 6. On October 5 the author's account published one post each on Threads and Instagram; the second posts of the day did not go out, the reason is below.

What I changed since the last entry. From about October 2 the server in the Netherlands that hosted the site was blocked at the Russian border in both directions: from Russia the site did not open without a VPN. On October 5 the site and the student database moved to a server in Moscow, and publishing posts and photos for Instagram and Threads moved to a server in Sweden: Meta's services are not reachable from Russia. The site opens on phone and computer without a VPN, and login by email and through Telegram works. I also fixed the mobile home page: the labels overlapped the model's portrait.

What it gave. There is no way to count how many people could not open the site from October 2 to 5: requests from Russia never reached the old server. The model has 84 followers on Threads, up from 83 on October 4, and 3 on Instagram. The author's account has 20 followers on Threads and 9 on Instagram.

Mistake of the day. From October 1 to 5, 9 of the model's 10 Instagram posts did not go out. The publishing approval in the spreadsheet was set, but the text column was empty, and the script silently skipped such rows. On October 5 two of the author's posts failed for the same reason. I learned about it not from the log but from the drop in reach. Now the server checks the rows for the next 24 hours every hour and messages me in Telegram, and at 23:30 it sends a report: what went out today and what is ready for tomorrow.

What I do tomorrow. On October 6 the model returns to Instagram: two photos, at 18:00 and 20:10. Six of the missed frames will go out one a day from October 11 to 16. In the evening I check by the report that all of the day's posts went out.

Want to try what I am testing right now? Start with the free lesson.

Day 22 / 90October 4, 2026

Day 22. Four days rewriting the model's rules and autoposting: the model gained 12 Threads followers, the author's account went quiet again

83 / 3
model's followers: Threads / Instagram (was 71 / 2)
≈870 ₽
spent Oct 1–4 (10 $); ≈17,300 ₽ since Sep 17
12
author's hours for Oct 1–4, three a day
0 ₽
earned

Days 19, 20 and 21, October 1 to 3: no entries, I am writing four days as one. On Day 18 I promised three things. The second frame of the model's test went out on October 1 at 19:30. I broke the promise not to generate the outfit reel: another 10 dollars went into it, and it never came out. The author's account posted twice a day until October 1, and from October 2 to 4 it published nothing.

What I changed since the last entry. On October 2 I rewrote the model's main document: the set of rules her posts and frames are made by. It is based on two outside reviews (audits) of her posts and audience, and I checked the document section by section. The model no longer has to end a post with a question to the reader; she now has her own opinions and topics: films, the city, travel. The model's open Telegram channel is published from the server on a schedule, like Threads. Frames for posts are made ahead from the spreadsheet rows: twelve are ready for October 5–10.

What it gave. The model has 83 followers on Threads, on September 30 it was 71. The best post, a photo by a café window on October 2, got 2,304 views against a median of 407 for October 1–3. On Instagram, not a single photo of the model has gone out since October 2. The "warm frame" test: the window frame of September 30 got 1,093 views, the empty-mug frame of October 1 got 311. I set follows per post as the criterion in advance, and only the app shows them, so there is no verdict on the test yet.

Mistake of the day. On October 2 none of the model's five scheduled Threads posts went out on schedule; two appeared outside the slots. The server was fine: the spreadsheet had no texts for that day. I was rewriting the rules and did not check what was due to go out the next day. The rules now have a deadline: the next day's texts are in the spreadsheet by 21:00 the day before.

What I do tomorrow. On October 5 I take the follows for the two test frames from the app and close the test. From October 6 to 19 the model runs a two-week check of the new rules. I am bringing the author's account back to one post a day.

Earlier days

Want to walk this path yourself?

Start with the free lesson: how the business model works and where the money in this niche comes from.