Google Search Topic Trends
Here's how I use Google Search, based on my search history since Jan 2021. For each topic, I've listed the number of chats and the average month-on-month growth.
| Topic | Count | Growth |
|---|---|---|
| Indian Celebrities and Directors | 1742.0 | -0.116 |
| JavaScript Libraries and DOM | 1613.0 | -0.727 |
| Singapore Local Info | 1479.0 | 0.200 |
| Python Programming Tools | 1402.0 | -0.238 |
| Node.js and Happy-DOM Issues | 1401.0 | -0.164 |
| GitHub and Git Tools | 1321.0 | -0.013 |
| D3 and DuckDB Data Visualization | 1321.0 | 0.235 |
| Testing and Code Tools | 1311.0 | 0.050 |
| Tamil Cinema Movies | 1170.0 | 0.169 |
| Browser and Web UI Tools | 1163.0 | 0.202 |
| OpenAI and AI Research | 1080.0 | 0.603 |
| India Data and Services | 1006.0 | -0.112 |
| Markdown and HTML Parsing | 972.0 | 0.094 |
| HTML and Bootstrap UI | 964.0 | -0.195 |
| Movies and Actors | 911.0 | -0.309 |
| AI Models and ChatGPT | 902.0 | 0.416 |
| Gramener Company Info | 848.0 | 0.009 |
| Pandas Data Analysis | 839.0 | -0.298 |
| Business and Corporate Reports | 817.0 | -0.243 |
| Email and Cloud Storage Services | 786.0 | 0.004 |
| Microsoft and Linux Utilities | 776.0 | 0.019 |
| SVG Graphics and Logos | 743.0 | -0.172 |
| APIs and Tokens | 691.0 | 0.289 |
| SQL and Databases | 682.0 | -0.049 |
| Mixed Tech and Media | 644.0 | 0.092 |
| Linux and Command Line | 622.0 | 0.029 |
| PowerPoint and Power BI Tools | 584.0 | -0.260 |
| Tech Devices and Support | 575.0 | -0.103 |
| LLM and Language Models | 566.0 | 0.331 |
| HTML Parsing and Beautification | 559.0 | -0.030 |
| HTTP and API Errors | 529.0 | 0.096 |
| Docker and Dev Containers | 527.0 | 0.015 |
| Google Services and Security | 524.0 | 0.090 |
| Data Visualization and Diagrams | 514.0 | -0.111 |
| Fonts and Text Embedding | 505.0 | 0.062 |
| JSON and Data Formatting | 485.0 | -0.039 |
| Time and Historical Data | 477.0 | -0.010 |
| CSS Styling and Web Design | 474.0 | -0.107 |
| Public Datasets and Data Science | 465.0 | -0.009 |
| Bangalore Local Info | 456.0 | -0.102 |
| Currency Conversion Rates | 453.0 | 0.036 |
| Unicode and Emoji Symbols | 407.0 | -0.025 |
| Media Conversion and Graphics | 388.0 | 0.015 |
| Minecraft and Gaming | 378.0 | -0.110 |
| Spreadsheets and Excel Functions | 351.0 | -0.105 |
| Books and Popular Culture | 334.0 | -0.033 |
| Chennai Local Info | 307.0 | 0.031 |
| Indian Culture and Movies | 189.0 | -0.024 |
| Education and Professionals | 182.0 | -0.046 |
| Graph Theory and Embeddings | 171.0 | 0.008 |
It didn't suprise me to see tech searches on top, but "Indian Celebrities and Directors" as #1 was a surprise. So was seeing salman khan, vijay antony, and vishal as my most searched celebrities.
I did nudge ChatGPT to:
Look closely at the numbers as well as the image. What insights can you draw from these? Aim for non-obvious non-trivial insights. Run correlations or any other analyses on the data to go deeper and come up with material suitable for a deep research paper.
What it shared was really insightful and actionable. Here are excerpts:
- AI research explodes fastest in absolute terms.
- Highest upward slope: OpenAIāÆ&āÆAI ResearchāÆ+0.60 searches/month.
- LLM & Language ModelsāÆ+0.33 searches/month
- APIs & TokensāÆ+0.29 searches/month
- Confirms āAI gravity wellā accelerating after ChatGPTās release.
- Suggests growing reliance on doc, spec, and tokenāauth guidance (API quotas, scopes).
- Outlier decline in Movies & Actors (ā0.31) vs. upswing in AI Models & ChatGPT (+0.42). My entertainment queries were cannibalised by AIāworkflows interest. Spare leisure time is now redirected to AI tinkering. (I've almost entirely stopped watching movies.)
- My search interests are diversifying. The HHI slopeāÆāāÆā1, indicating that curiosity is spreading over more niches. (Good to know!)
... but the really powerful (and actionable š”) insights were:
- PCA shows three āpersonasā.
- 32āÆ%: classical dev stack (JS DOM, HTMLāÆUI, Python Tools)
- 24āÆ%: AIābuilder stack (OpenAI, LLMs, APIs & Tokens)
- 16āÆ%: India/Singapore geoācultural info (Celebs, Local Info, Tamil Cinema).
- š” Segment sharing & learning along these axes (e.g. separate newsletters or dashboards.)
- Testing & Deployment topics sit at the correlation crossroads.
- Testing &āÆCode Tools correlates strongly with both dev stack and AI stack clusters (Ļ>0.7). Serves as integration glue.
- š” Invest in testāautomation expertise for outsized ROI. Improvement here amplifies throughput across all other areas.
- Volatility is domaināspecific. Based on Coefficient of Variation:
- Highest CV: SQLāÆ&āÆDatabases (0.24), TestingāÆ&āÆCode Tools (0.23).
- Lowest CV: Currency Conversion Rates (0.05), Singapore/Bangalore Local Info (0.06).
- Realātime āfireāfightā topics are episodic; stable reference needs are perennial.
- š” Automation opportunity: cache common reference lookups locally; allocate ādeepāfocus blocksā for debugging spikes.
I would not have thought of these analyses possibilities!
Process
To extract Google Search data, go to Google Takeout and export "My Activity". You'll be emailed a .zip file. Pull out Takeout/My Activity/Search/MyActivity.json as google.json.
Extract all the search terms by extracting them from all lines with "title": "Searched for ...":
grep 'Searched for' google.json | cut -c26- | sed 's/",$//' > google.txt
Run topicmodel to identify topics:
export OPENAI_API_KEY=...
uvx topicmodel google.txt --ntopics 50 --nsamples 30 --output topics.txt
Merge topics.txt into google.json using merge.py:
uv run merge.py
ChatGPT then creates the charts and analyses.
Comic Story
I created a comic story (PDF, 22MB) with my PicBook tool using my photo and this prompt:
- I analyzed 4 years of my Google search history. [Draw: Night study room. Protagonist unrolls a dusty scroll from a trunk labeled āBACKUPSā. Ghostly numbers and topic names swirl out. Candlelight + laptop glow mix.]
- It's mostly tech. That was no surprise. [Draw: Protagonist bored, leaning on a pile of hefty tomes: āJS DOM (1613)ā, āPython Tools (1402)ā. Heās sipping chai, half-asleep. A speech bubble with āmehā.]
- Butā¦ Indian celeb searches beat JavaScript/Python! [Draw: Two weighing scales. Left: glossy Bollywood headshots (Salman Khan, Vijay Antony, Vishal caricatures). Right: code books (āJavaScriptā, āPythonā). Celeb pan heavier. Protagonistās jaw drops, eyebrows sky-high.]
- I asked ChatGPT to analyze the data. [Draw: Protagonist points like a commander at a blue glowing genie emerging from a chat window labeled āChatGPTā. The genie wears a lab coat, carries a calculator and PCA plot.]
- AI research is definitely the fastest growing category. [Draw: Rocket labeled āOpenAI & AI Research +0.60ā blasting off. Smaller booster rockets āLLM Models +0.33ā and āAPIs & Tokens +0.29ā follow. Protagonist clings to a ladder on the main rocket, hair blown back.]
- It increased at the expense of movie searches. My leisure time's shrinking! [Draw: A sinking film reel labeled āMovies & Actors ā0.31ā while a hot-air balloon āAI Models +0.42ā ascends. Protagonist drops popcorn to grab a laptop mid-air.]
- My searches are growing more diverse over time, though. [Draw: A hillside with a sign āHHIā. Protagonist sleds down happily, scattering topic signboards into many smaller paths. An arrow āDiversity āā points outward.]
- The searches reflect 3 personas: Dev, AIābuilder, Geoāculture. [Draw: Pie chart shields (32%, 24%, 16%) held by three avatars of the protagonist: (1) Devātoolbelt of JS/HTML/Python. (2) AIābuilderāAPI keys, tokens, model cards. (3) Geoāculture fanāmap of India/Singapore, film posters. They stand back-to-back heroically.]
- Testing as a profitable area of focus. It correlates with TWO of those personas. [Draw: A busy intersection signpost: āDev Stack ā AI Stackā. In the middle, a tollbooth labeled āTesting & Code Toolsā. Protagonist hands a āUnit Testā ticket; lanes merge smoothly.]
- I didn't think of these insights. ChatGPT is a better data scientist than me! [Draw: The protagonist is solving a simple puzzle while the blue glowing genie labelled ChatGPT is solving a FAR more complex one.]