What I Learned From Live-Tweeting My Google Searches

Last May, I wrote a script that would log into my Google history page, find my most recent search queries, and tweet them from my public Twitter account. It tweeted searches from anywhere I was logged in with Google, including my phone, in real time.

I kept the script running continuously for over a year, generating about 17k tweets. A few nights ago, for various reasons, I decided to turn it off, and commented out the cron job that ran the script.

Frankly, I miss it.

The feed changed my habits and relationships in ways that I’m only starting to appreciate now that the code isn’t running, isn’t a part of me anymore. I think it was an interesting experiment, and I’m going to try to summarize some of my conclusions. My friend Theo Thimo, who originally had the idea for this project, still has his feed running, if you’d like to check it out in action.

How It Was Useful

The main reason I miss my Google search feed bot is that it was actually quite useful. It worked something like a Slack integration! My friends and coworkers were ambiently aware of the research I was doing, the questions I had, in real time. This helped us sync up and discuss important things. Now I have to remember to explicitly tell them what I’m working on.

Another plus: one of the well-known problems with Google is that it doesn’t do a good job of addressing your “unknown unknowns”. Quite a few times, I’d be Googling myself into a dead end, and someone on Twitter would point out a Better Way.

Finally, the bot also helped me make new connections. Google searches are essentially keywords, and this meant that my Twitter feed became a stream of keywords relevant to me. Unsurprisingly, this helped people (actually mostly bots) find me. I have effortlessly made a number of good friends and acquaintances over the past year thanks to this script.

What I Hid

Did I have a way to hide certain queries? Absolutely. If I used an incognito tab, I could search without it showing up in my history page and therefore also my Twitter feed.

In practice, I didn’t do this very much. When I did “incognito tab” something, it was usually for one of three reasons:

  • It was related to my family. There’s something deeply weird about Googling one’s mom or dad, even weirder than Googling oneself.

  • It was related to a person or organization that I was in a competitive or vulnerable relationship with. For example, a few weeks after the script started running, I graduated from college and started to half-heartedly look for a job. When researching companies, I often “incognito tabbed”, due to something between embarrassment and self-interest.

  • Irrelevant, repeated queries. Over the course of the experiment I gained about 600 followers, many who don’t care about programming. Occasionally I took mercy on these people by opening up an incognito tab when debugging a webpack.config.js. I didn’t want to clog the feed.

I think these three categories point to three important use-cases of privacy: 1) a way to care for the emotions of others, 2) a tool in competition, 3) a filter of irrelevant information.

How It Affected Me, And Why I Turned It Off

I gained a lot of interesting reflexes over the past year. For example, even though the script isn’t running anymore, I’ve instinctively opened up an incognito tab for a worrisome query. When I watch someone else Googling, I often feel a bit anxious for them if they’re searching someone’s name. I think this suggests that I’ve sort of adapted to a low-privacy environment.

I think I also got accustomed to a constant stream of fav’s and re-tweets on Twitter, thanks to my “seamless content”. I was rewarded for searching about relevant topics, or going to relevant places. According to Twitter Analytics, my most popular tweet in December 2015 was simply the address of a party I was heading to:

Ultimately, it was concerns about laziness that made me turn off the bot. The bot was rewarding me for keywords, rather than coherent thoughts. Looking back on my tweets from the last year is sort of bleak: all it really tells you is that I went places in Brooklyn, wrote software using Ruby, C#, and JavaScript, and made shallow inquiries into a few other subjects. I know that more was going in my head and heart during that time, and I want Twitter to force me to put that stuff into a concise form.

Technical Thoughts

My friend Theo Thimo came to me with the idea: he thought of it as “seamless content”, an effortless form of writing. He asked if I could build it, and of course I said yes. “I’m sure I can just hit some Google API!” I thought. I told him I’d do it for $20. (He’s a friend.)

Building google-search-twitter ended up being a learning experience, because Google doesn’t provide an API for search history data, despite the fact that they’ve been quite busy with the history.google.com page, adding location data and voice queries.

Some of the fun technical challenges I faced in getting the script to work:

  • Installing an X window server on a t2.micro instance in order to get around Google’s well-meaning security measures. Unsurprisingly Google locked down my account the first time I tried to log into it via browser automation from an ec2 instance in Oregon. By manually going through a two-factor authentication process on the box, I was able to get Google to accept the state of affairs. I later automated this.

  • About three weeks after I finished my initial implementation, Google switched the history page to an Angular single page app. It was originally a server-side generated page controlled by query string parameters, the type of easily-scrapable webpage that is essentially its own API. I ended up writing a god-awful regex to parse the search queries out protocol buffers bootstrapped into the page. I’m sure there’s a better solution. This was a teachable moment about the dangers of relying on “unofficial APIs” and the benefits of server-side templating.

  • Occasionally Google would add new stuff to the history page, and my Twitter feed would start spewing it out. For example, at one point Google added app usage to the history page, and my Twitter feed became mostly “Used One App: Facebook”, “Used Three Apps: Twitter, Harvest, Strava”. The same thing later happened with voice queries. These things were all opt-out-able. It was good to be conscious of these things, in my opinion.

Patrick Steadman
Thoughts or questions? Chat now.