Catching AI names before the keyword tools do

What it does

Every day it scans the places where a new AI name first appears — posts from X accounts you follow, Hugging Face’s trending list, GitHub repos that suddenly take off, and sitemaps of sites you track — and produces a report of terms that look like names of new things rather than common words. Each name also gets a keyword table: the generic search phrases behind it (“ai image detector” behind “SynthID Detector”), sized and given a prioritize/watch/skip decision. Fourteen days later, each lead is checked against Google Trends to score the hit rate per signal.

The interesting bit

The premise is a lag map: keyword tools read step 5 of a name’s life (Google Trends), while this reads steps 1–4 — the X post, the HF trending entry, the one repo gaining 800 stars a day. The README’s founding example: a repo created on 2026-09-24 hit 7,700 stars in nine days, people searched its name from day two, and no keyword tool listed it. The self-check loop is the honest part — it measures whether its own signals actually predicted a rise, per signal.

Key highlights

• Free signals (Hugging Face, GitHub breakouts, sitemaps, Search Console) run without any API key; the rest go through one paid AIsa key, roughly $15/month at the example config
• GitHub breakout rule: repo under 14 days old averaging 200+ stars/day, plus the search variants people actually type (“niko strata”)
• A small model judges each term: name of a specific thing, or common word — common words get listed at the bottom, not buried
• Keyword table pairs each new name with the durable generic demand it serves, since names are short-lived keywords and the job they do is what people keep searching
• Standard library only, Python 3.10+; runs on GitHub Actions

Caveats

• Costs money per call via AIsa (X reads, Trends, domain data, the judging models); without the key, the report lists every term unsorted
• The optional GoAnyAPI integration is young — in the author’s tests, calls took 20–55 seconds and about one in ten failed
• The README is candid about limits: Google Trends’ last two days are unreliable (the code now drops them), Ads volumes lag ~2 months, and the 14-day review of generic terms has no results yet

Verdict

Worth a look if you build SEO pages, newsletters, or directories around new AI tools and want the name the week it appears. If you don’t chase emerging terms, it’s a well-documented curiosity — though the signal-lag framing is instructive even for non-SEOs.