Last week I put 15 AI tools on one Google Trends scale. The obvious objection: those were search terms. "Gemini" includes the zodiac sign, "Claude" includes people named Claude, and "cursor" is mostly the thing you move with a mouse.
Google Trends has a fix for this: topics. A topic is an entity in Google's Knowledge Graph (the AI company Runway, as opposed to a fashion runway), and its interest counts the searches Google attributes to that entity across spellings and languages. So I re-ran the same 15 tools as topics and compared them with the term results.
What changed
Both runs are on a scale where 100 is ChatGPT's highest week, so each bar is "topic interest as a share of term interest, relative to ChatGPT".
- Cursor drops to 8%. As a term it looked as big as NotebookLM. As a topic it is the smallest tool on the list. Nearly all "cursor" searches are about the mouse pointer.
- Runway drops to 29%. Fashion runways and airport runways.
- Gemini keeps 68%. Bigger than I expected. The zodiac sign is a real share, but most "gemini" searches are now the AI.
- NotebookLM keeps 91% and Midjourney 82%. Made-up names are almost all signal.
- Most tools land at 55-70%. That is roughly the share of the term that belongs to the tool's topic. The rest goes to related topics, spelling variants Google files elsewhere, and noise.
The ranking itself barely moves at the top (ChatGPT, Gemini, Claude), but the long tail reshuffles. Cursor falls from 9th to last.
| Tool | As a search term | As a topic |
|---|---|---|
| ChatGPT | 83.0 | 83.6 |
| Gemini | 45.9 | 31.2 |
| Claude | 15.6 | 9.5 |
| Copilot | 4.5 | 2.7 |
| Grok | 3.6 | 2.3 |
| DeepSeek | 3.0 | 2.2 |
| NotebookLM | 1.8 | 1.6 |
| Perplexity | 2.0 | 1.3 |
| Character AI | 1.4 | 0.8 |
| Suno | 1.3 | 0.8 |
| ElevenLabs | 1.0 | 0.5 |
| Grammarly | 0.8 | 0.5 |
| Runway | 1.1 | 0.3 |
| Midjourney | 0.3 | 0.3 |
| Cursor | 1.8 | 0.2 |
Average interest, worldwide, last 12 months; 100 = ChatGPT's peak week.
How to get topic IDs
Topics have IDs like /g/11khcfz0y2 (ChatGPT) or /m/0gg4qg3 (Grammarly). Google Trends' autocomplete endpoint returns them for any term:
https://trends.google.com/trends/api/autocomplete/cursor?hl=en-US
The response starts with )]}' (an anti-JSON-hijacking prefix); strip that line and you get JSON. For "cursor" the top suggestions are:
| mid | title | type |
|---|---|---|
| /g/11x21hly6v | Cursor | Software |
| /g/11xs8l0p90 | Cursor | AI company |
| /m/06mfbd | Cursor | User interface |
You then pass the mid anywhere you would pass a keyword: on the Trends website (?q=/g/11x21hly6v), in pytrends, or in your own requests.
Two things I learned picking them:
- Pick by type, not just title. "gemini" returns three topics called Gemini (Software, Topic and Language model) plus the zodiac sign. "perplexity" returns both "Perplexity" (a generic topic) and "Perplexity AI" (the company). Choose the one that matches what you mean.
- Names change. "notebooklm" now resolves to a topic titled Gemini Notebook, so it looks like Google has renamed it. If you look topics up by title, an exact-match search for "NotebookLM" finds nothing.
Caveats
- Topic interest depends on Google's own classification, which you can't see or audit.
- A product can have several topics (Cursor has a Software topic and an AI company topic). I used one per tool, so a tool split across two topics is understated.
- These are two separate measurements joined through ChatGPT. That works because ChatGPT's term and topic agree almost exactly (83.0 vs 83.6), but it is an assumption.
Doing it without the lookup work
I built an Apify Actor that does both steps: Google Trends Scraper. Disclosure: it's mine, and it's pay per result ($1-2 per 1,000 rows; empty results are free). Set resolveTopics: true and it looks each keyword up as a topic (exact title match first, otherwise the top suggestion), then stitches everything onto one shared scale. Or pass mids like /g/11x21hly6v directly for full control, which is what I did here.
{
"keywords": ["chatgpt", "gemini", "claude", "cursor", "runway"],
"resolveTopics": true,
"timeRange": "today 12-m"
}
Each result row includes the topic it resolved to, so you can check that "cursor" meant the editor and not the pointer.
What would you measure as topics instead of terms? I'm curious which names are the most polluted. My guesses would be "Apple", "Amazon" and "Python".

Top comments (1)
We need to output a short YouTube comment, casual, one or two sentences, starts with a specific reaction or question about this video. Not generic praise. Avoid certain phrases. Must be casual, maybe like "lol, why only 8%?" or "I wonder why cursor is so low". Should be about the video content: only 8% of "cursor" searches are about Cursor. So comment like "so why are so few searches actually about the tool?" Could ask about methodology: "did you filter out brand