Intelligence of bloggers and trends for PointGPT
A database of AI creators across five sites, a swipe file of viral Threads posts, analysis of competitors and a ready-made content engine. Collected via API and browser - a reproducible pipeline.
PointGPT - neural network aggregator in one subscription without VPN
Key figures
Problem
Find growth points for PointGPT through bloggers and content: collect the database of AI creators into a table, identify trends for transfer to the content plant and for integration into the service, parse Instagram of competitors and parse what flies into Threads.
What did you collect?
Telegram
40AI channels ≥2000 subscribers (creators, content factories, AI news). Trifles, media exchanges, and crypto-fakes are strictly eliminated. Coverage, view-rate, contact for each.
only strong AI creators (≥3000 subscribers + clear AI focus). Out of 320, 15 were selected - without 0-accounts and micro. Top - @zavisual 446k.
TikTok
41AI creators ≥5000 subscribers. Removed neurographics/psychology and crypto. Top - @rileybrown.ai 637k.
YouTube
47AI channels ≥10,000 subscribers: Matt Wolfe, The AI Advantage, VideoSmile, Wes Roth.
X / Twitter
17curated list of AI accounts for Western trendwatch.
Threads
60reassembled the previous conclusion: the niche is alive. Using a logged-in browser, I collected 60 viral AI posts from the “Top” tab (6-14 thousand likes), pulled out the hook format and adaptation for PointGPT from each. Plus a separate database for 66 creators.
Threads: reassembled the output - the niche is alive
At first, auto-collection via API/Apify yielded few posts, and the conclusion was “the niche is empty.” I double-checked it with my hands and it turned out to be a limitation of the tool, not a reality.
Downloaded GitHub skills for collection
Scweet (X, filter by likes = only flown in) and threads-api - installed and checked. threads-api turned out to be outdated (archive 2023, endpoint is dead), Apify hit the monthly limit.
Collected live via browser
I went to Threads, through the “Top” tab, using 6 keys, I made 60 viral AI posts with metrics - likes, reposts, forwards. Only what really caught the activity.
Hook swipe file
I defined the format for each post (prompt distribution, list, pain→solution, verdict-hype, interactive) and described the ready-made adaptation for PointGPT with a ref link in the pin.
Base of 66 creators
From the “Profiles” tab + authors, Viralov collected a database for outreach: link, subscribers, views, funnel, connection status. Top - rinat__ai 25k, mr.pr0mpt 10k, kate.shabaeva 8.7k.
Content engine
7-day posting plan, 6 ready-made posts and DM templates by segment - the table immediately turns into actions.
Insight that is more expensive than collection
The “don’t pay for a bunch of subscriptions - all in one” angle is going viral on Threads in all languages. This is literally PointGPT’s offer - the No. 1 format to work with.
Trends that convert
Each is marked along axes: what to pick up for creators, what to build into the service, what to put into the newsletter and target.
Put yourself in the film
Kling and Nano Banana. One of the most monetized formats among AI creators.
Reviving old photos
Strongest emotion, stable reposts to personal networks.
Prompt "don't change your face"
Factory format for 190k subscribers. Zero entry threshold.
Content factory on Sora 2
The record for reposts in the sample is 642. A dream topic for SMM.
Neural networks without VPN
The pain of the Russian market. This is the core of PointGPT's own offer.
Earning money from AI content
Money is a hook and a tone standard for founder posts.
I checked Instagram with 30+ AI services. In the selection of 15, twelve are active on the site - from @syntx_global (10k) and @ideogram.ai (21k) to @photoroom (1.9M) and @picsart (4.2M). A ready-made playbook that can be copied.
- —Photoroom (1.9M) and PicsArt (4.2M) rely on the “before/after photo” format and AI effects - these are almost our WB/Ozon cards. Direct content reference.
- —Higgsfield and Study24 have a showcase of creators (@*.creators): reposting user works for a tag - a content flow and a growth loop.
- —And Russian-language competitor aggregators (ChadGPT, BotHub, GoGPT) are on Telegram and do not operate Instagram - there is more space in the RU segment.
- —Recipe for PointGPT: @pointgpt.creators + Reels “repeat the trend” + prompt carousels + UTM for all links.
What we sell from our competitors’ Instagrams
I pulled out work directions from those who run Instagram and transferred them to PointGPT.
Showcase of creators @pointgpt.creators
who has: Higgsfield, Study24
Reposting user works for a tag - free content and a growth loop.
Photo "before/after"
who has: Photoroom 1.9M, PicsArt 4.2M
Carousel “was → became”, right under the WB/Ozon cards.
Reels “repeat a trend in a minute”
who has: almost none in RU
Tutorials for the trend - occupy the empty field first.
Carousels of ready-made prompts
who has: facebox 190k, promtyi 119k
1 carousel = 5-7 prompts + “repeat in bot” button.
Cinematic video demos
who has: Luma 510k, Runway 458k, Pika
Kling/Sora/Veo demos - showing multimodality.
UTM for everything
who has: Study24
We measure IG traffic → bot, disable what does not convert.
How it's put together
- 01I created a reproducible pipeline on Node.js: each collection is repeated by one command, the cache saves limits, the data is immediately classified.
- 02Telegram collected directly through MTProto, other platforms - through Apify actors. Sources are easily expanded to new niches.
- 03For each channel, I pulled out not only subscribers, but also reach, view-rate, posting frequency, contact from the description and type (creator / content factory / newsmaker).
- 0413 trends from top posts are marked along three axes: what to lure into the content factory, what to build into the service, what to pick up in the newsletter and target.
- 05I ran everything through a strict AI cleaning (audience threshold + checking the relevance of each): out of ~700 collected, I left 160 that were really worthwhile - without 0 accounts, micro and offtopic.
- 06Downloaded and configured GitHub skills for viral collection: Scweet (X, filter by likes) and threads-api. For Threads, collection through the browser turned out to be working - threads-api is outdated, Apify has hit the limit.
- 07I reassembled the output from Threads: I collected 60 viral AI posts (likes up to 14 thousand) and a database of 66 creators with subscribers - the niche is alive, the previous “empty” output was a limitation of the tool.
- 08For outreach, I collected 40 targets across 5 platforms + a separate database of 66 Threads creators with DM templates by segment; for publication - swipe file formats, 7-day plan and ready-made posts.
Tools
What's in the table
15 sheets, about 460 lines - after hard cleaning, only the relevant remains, with filters, priorities and contacts. The Threads swipe file and the creators database are in separate sheets.
Do you need the same reconnaissance or an AI pipeline for a task?
I collect databases of creators, swipe files and reproducible data collection pipelines for a specific product. Let's discuss your problem - 30 minutes, free.