Evidence-based SEO

I start with measurement infrastructure. Without clean data from [Google Search](/seo/) Console and Google Analytics, every claim is guesswork. Verify that your sitemap is submitted, canonical tags are correct, and crawl controls (robots.txt, noindex) are set intentionally. This is the boring ground that makes everything else credible.

Next I establish a baseline. Pick a 90-day window before any change. Pull organic traffic, average [SEO Ranking](/seo-ranking/) positions for your target keywords, and, critically, conversion rates — sign-ups, sales, or whatever matters. I once skipped this and spent three weeks 'improving' a page only to realise seasonal traffic was the real driver.

Then I build a prioritisation matrix. Plot your pages or topics on two axes: opportunity (search volume, current impressions, gap to top 3) versus difficulty (domain authority, competitor strength, technical constraints). The pages in the high-opportunity, low-difficulty quadrant get tested first. For example, a product page ranking #8 with 10,000 monthly impressions but a 2% click-through rate is a screaming opportunity. I rewrote its meta description to include a price promise and added a bulleted list in the snippet. The click-through rate went from 2% to 4.5% in two weeks. Traffic rose by 30%.

But traffic alone is not enough. I also tracked conversions. That same page saw no uplift in purchases because the content below the fold was still weak. That taught me to measure conversions alongside click-through. Always measure the metric that pays the bills, not just the vanity one.

Finally, I implement one change at a time, measure for at least four weeks (two for traffic signals, two for conversion signals), and record the result in a log. Over a few cycles this log becomes the best [SEO Best Practices](/seo-best-practices/) list you will ever own because it is specific to your site, your audience, your context.

Don't treat rankings as the only success metric. I have seen a keyword jump from #5 to #2 while organic traffic dropped because Google introduced a featured snippet above the organic results. Rankings without traffic or conversions can mislead you into celebrating a hollow win. Always triangulate: position, clicks, and conversions.

Don't skip the baseline. I once made a batch of optimisations and saw traffic rise 20% month-on-month. I thought I was clever. Then I plotted the same period from the previous year and saw the exact same uplift — it was seasonal. Without a baseline, you cannot attribute any change. Pull at least 90 days of historical data before touching anything.

Don't copy competitors blindly. A competitor might rank #1 for a high-volume term because their domain authority is 70 while yours is 25. Their page has 500 backlinks from .edu sites. Replicating their content won't move your needle if you lack the link profile. I wasted four months on that exact mistake. Now I ask: 'What is the evidence that this tactic will work on my site?' If the competitor's advantage is structural (authority, budget, brand), find a different opportunity.

Be wary of small sample sizes. If your test page gets 50 visitors a week and you see a 10% lift, that is noise. Wait until you have at least a few hundred data points per variant. For low-traffic pages, consider running a longer test or using a pre/post design with a control page that gets no change.

Watch for Google updates during your test period. In November 2023 I was testing a content expansion when the Helpful Content Update hit. My test page tanked, but so did my control page. I had to restart the test after the update settled. Mark on a calendar any expected update windows and extend your measurement period to account for turbulence.

I used to think evidence-based SEO meant running complex A/B experiments with dedicated tools like Google Optimise or VWO. It turns out the core workflow is much simpler: set a baseline, make one change, measure, repeat. The fancy tools help, but a spreadsheet and Search Console cover 80% of what you need. I have run perfectly good tests just by tracking week-over-week click share for a test page against a similar control page.

I also wasted months chasing competitor backlinks. I would find a page that ranked well, scrape their link profile, and start emailing the same domains. It rarely worked because those links were earned through relationships or content depth I did not have. Now I focus on linkable assets (original data, tools, guides) that naturally attract links, and I measure the impact on search visibility over time.

Another mistake: I treated raw traffic as the only success signal. I would celebrate a 50% traffic spike without checking whether those visitors took the action I wanted. One test increased traffic 40% but conversions dropped 10% because the new visitors were less qualified. The net effect was negative for revenue. Now I always set a primary and secondary metric before any test: primary is conversions, secondary is traffic.

I changed my mind on what makes a good first test. I used to pick pages with the highest traffic and 'optimise' them. Now I pick pages with the highest untapped potential — high impressions, low click-through rate, low conversion rate. Those pages give you the biggest signal for the smallest effort. The [SEO Checklist](/seo-checklist/) I use now starts with impression data, not traffic data.

I still struggle with incrementality measurement for content changes. If I expand a page, the ranking lift might come from freshness rather than the new content itself. I have not found a perfect way to isolate that. My current workaround is to compare the test page against a similar page that I leave untouched during the same period. It is not airtight, but it is better than nothing.