This post is one of a series on my content production system, developed over 10 years and designed to scale content end-to-end in 6 months as an acquisition channel.
Here are the other posts in the series:
- Why the winner takes all in Content Marketing
- What everyone gets wrong about Content Strategy
- Building Your Keyword Universe
- Minimum Viable Technical SEO Audit
- Recruiting good Freelance Writers
- How to write blog content at scale
- Growth engineering for SEO (this post)
- Link building for authority
If you really want to scale content, you can’t do it by human effort alone - you need engineering resource. There are two major strategies for generating content at scale for SEO: user generated and machine generated content.
A. User Generated Content
UGC is the easiest to build (from a technical perspective) but it can be very hard to get traction. Fundamentally you need to find a creative way to convince users to do data entry for you. If you can get them to submit content to the site, and structure that data so it ranks, you’ll be able to explode your SEO. For example, Thumbtack asks questions like “what makes your service stand out” during onboarding, and uses the answers on profile pages.
B. Machine Generated Content
It’s hard to motivate users to do data entry for you, so many go the machine generated route: use web scraping or build a unique dataset and display the structured data as pages. At Candor we built a site with all 11k+ recruiters who work in tech, and started ranking when people searched recruiter’s names. Tools like GPT-3 are opening up more possibilities in this area. However you can start small, with simple rules that enrich your existing content with target keywords.
A/B Testing SEO
It can be incredibly difficult to be scientific in content marketing, because it’s impossible to run a controlled experiment. Google serves personalised search results for every user, and we have severely limited control over what is shown. They also don’t like it when you serve different content on a page to different users (cloaking). This is one more reason why the winner takes all in this market: if you get enough traffic to run statistically significant tests, you can learn faster than the competition.
A. Correlation Analysis
When you don’t have enough data for scientific testing, you need to look at correlations and make some educated guesses at what to do. You can look at your own data around what seems to rank highly on your site or in your industry, but it probably makes sense to look at overall ranking factor studies for the big picture.
B. Trend Analysis
It might not feel very scientific, but switching something on or off then looking at the trend is one very common and effective way to estimate the impact on SEO. If something really worked you’ll see a huge spike, and you won’t really need statistical significance to prove the case. It’s important to keep notes on when changes were made in order to go back and annotate your analytics charts and see where the timing lined up.
Going one step further with this method: it can help to establish the existing seasonal trend in order to do a proper analysis. You don’t have to run a full causal effect model, but even something as simple as overlaying the average hourly traffic before and after can be enough to tease out the effect of your change.
C. Split Testing
Although we can’t truly scientifically A/B test with SEO, we can get close enough. The method was developed at Pinterest and later productised by Search Pilot, and it consists of splitting your pages by type of page, then making changes to 50% of the pages in that category. For example adding a ‘related interests’ bar to your Board pages to see if that drives more traffic.