Optimizing highspeedinternet.com Part 2

Content is king.

Like I mentioned in the last post, the city pages on highspeedinternet.com were pretty sparse. Here’s an example of what they looked like: https://web.archive.org/web/20160709211328/https://www.highspeedinternet.com/ca/los-angeles. Since the biggest site I’d worked on at the time was about 100 pages, my main approach to SEO was content. If you wanted a page to rank, and it wasn’t, you’d optimize the content. 

Nothing to optimize.

However, on HSI city pages in spring of 2017, the optimization was sort of “done” before you started. All the classic places to optimize were filled in dynamically. We’d done the meta descriptions a few months before, and messed around with title tags a bit. We added “best” and “top 5” but really only got marginal click through rate gains. Nothing substantial. There were zero paragraphs of actual copy. Just dynamically generated provider availability data layered onto city data.

Optimize for the user. 

Since there were about 30,000 city pages, my usual approach of analyzing copy, providing recommendations for a copywriter, and increasing value to the user would be a complete non-starter. (Or so I thought). Besides, as you can see from the Wayback Machine, there wasn’t even any place to have a writer update anything. The only human-generated content on any city page was the user reviews of different providers, and this had an extremely negative sentiment. In fact, so much so, that some of the clicks coming into the site were things like “comcast sucks”. Not a ton, but a click here and there, enough to register in GSC. This got me thinking… if user reviews were getting picked up by Google, and the site was actually ranking and getting clicks for this stuff, then maybe there was some potential for swaying that sentiment back to positive with our own human generated content. 

Content Planning.

I don’t remember how the conversation exactly went, but I started talking with Paul the SEO director, and Carson the manager on HSI, and asking for a way to add content to the city pages. I wanted real human written content as a test to see if we could get movement on a few pages. It turns out that Paul had a friend, Mike, who managed an offshore content generation company. We got in touch with Mike and his rate was low enough that we could create content for a few hundred pages for relatively cheap. Keep in mind that highspeedinternet only generated about $200k-300k in revenue for the year, so there wasn’t a ton of budget for this.

Write write write.

We settled on the following plan: we’d have Mike’s team write about 100-150 words for each page that would get copy. It was important to keep it short for a couple reasons: first Mike charged by length and shorter was cheaper. But also, I wanted it short enough that we weren’t producing too much “fluff”. My goal was that Google would see something valuable. In order to ensure there wasn’t too much overlap, and to keep the content locally focused, we had the writers keep the topic to a description of the area of availability for the top internet providers featured on each page. Every city page had a dynamically generated map that overlaid a provider availability footprint on an image of the city. By having the writers describe this footprint, we ensured that each piece of copy was different (since the shape of every ISP footprint is different), and unique to the city where it would live (since it used place names and streets from that city). This is what a typical city page looked like after creating the copy.

Scope and results.

The first batch of copy was the top 70 city pages. The writers wrote their two paragraphs of copy and we published it on Los Angeles, New York, Phoenix, and the rest of the top 70 populated cities in the US and held our breath. Well at least I did. At first nothing happened. I checked the click through rate, rankings, and traffic for those top 70 pages against the rest of the 30,000 in the GEOset and it was pretty underwhelming. Finally after a week or so, it looked like at least they weren’t worse. This was mid February, which is the bottom of a trough for seasonal volume, so if traffic started to grow around this time, there’s no real reason it has to be because of rankings. 

The best hail mary guess of my career to date was what I did next. I suspected that city pages might be more or less like fish in a school, and perhaps Google was evaluating them together. After all, they did rank for the same exact keyword. Because of this I suspected that maybe there was some sort of “critical mass” needed, and that 70 pages just wasn’t enough to matter. Two tenths of a percent is pretty small. Anyway, I suggested that we just continue writing. If 70 pages had tentative, almost imperceptible movement, maybe 300 would work. So we had the freelancer write 230 more, to fill out the top 300.

The train starts moving.

Once the 300 were live, the results were clear and positive. The pages with copy were up between 10-100% more than the rest. Keep in mind that “the rest” were getting maybe a session a day at most, so 100% growth was only 10 sessions. But to me the data was clear. The 300 pages with copy were ranking slightly higher than the rest, and enough higher to produce real traffic growth. I suggested that we keep pushing, and we had the writer do another batch of 900 this time. The reason the batches had to keep getting bigger is because the city pages they would be published on represented smaller and smaller cities. Population from the largest city in the US (NYC) to the smallest (at the time, Cucumber, WV, with 30 people) drops off roughly exponentially. This means that for copy batches to reach equivalent amounts of potential users, the number of pages in each batch must increase exponentially, or at least approximately so. 

Liftoff!

Once publishing for the 900 was underway, Google couldn’t ignore us anymore. I’m writing this on March 9th, 2026, which is funny because March 9th, 2017 is the last day highspeedinternet.com was under 1000 organic sessions. It wasn’t overnight but it did catch like wildfire. Before all the content was live, the city pages ranked about position 30 for “internet providers”. Somewhere mid-way through publishing the batch of 900 the Los Angeles page ranked number one on Google for “internet providers”, a position it would hold consistently until 2023. The rest of the city pages followed. As they rose in rankings, and increased in traffic, we kept publishing. Because the increase in traffic massively boosted revenue which we had not budgeted for, there was plenty of funding to continue writing. Over the course of the next year, Mike’s team wrote copy for every single one of the 30,000 city pages on the site. 

Ranking details and timing.

Interestingly, the increase in rankings didn’t exactly coincide with the publishing of content. That is, when 70 were published, there was a shiver at best. When 300 were published, things started moving, but moderately and only on the pages with copy. BUT when 1200 were published, ALL 30,000 started moving. Even though 1200 is only 4% of 30,000, the traffic the 1200 represented was probably closer to 30% or more of the GEOset. Remember, we did the biggest cities first. This indicates something extremely interesting and insightful about GEOsets: they function somewhat like a sideways page. That is, if you can rank enough of them, you can rank ALL of them. But even though they rank somewhat together they don’t rank in exact lockstep. The best way to think of GEOsets is like a slinky, where if you move enough of one end, the rest will move, but only eventually.

It also proved out the validity of natural granularity of GEOsets, and that internet providers’ natural granularity is at the city level. If we had tried to to individual copy on zip pages, or state pages this probably wouldn’t have worked. Zips would have been too similar, and cannibalized each other, and states would have been too few, and not specific enough for any one person. 

Anyway, traffic grew about 10-20% week over week from March 2017 throughout the rest of the year. This astronomical and steady growth would eventually lead to catastrophe, which I’ll go into later.