The SEO Experiment That Made Grammar Pages Rank 2x Faster
On February 27, 2026 I ran a controlled SEO enrichment test on 14 grammar exercise pages. By March 15 they were improving about twice as fast as the rest of the site, which was enough to roll the pattern out everywhere.
At the end of February I ran a very unglamorous SEO experiment on english-exercises.org.
The hub pages were doing alright. The individual exercise pages mostly were not. Pages like /first-conditional-exercises and /zero-conditional-exercises had real search demand, but a lot of them were stuck on page 6 or page 7.
When I checked what Google could actually see, the answer was mildly embarrassing. Most sub-exercise pages had about 15 words of server-rendered HTML, a hidden sr-only H1, and no structured data at all. Competitors were giving search engines 400 to 1,400 words, visible headings, FAQs, examples, and cleaner signals about what the page was for. We were basically saying “trust me bro” in HTML form.
What The Pages Looked Like Before
The audit on February 26, 2026 made the problem pretty obvious:
- 582 pages had little more than a hidden H1 as server-rendered content
- 485 pages had zero JSON-LD structured data
- 249 pages were getting impressions but no clicks at all
first-conditional-exerciseshad 217 impressions and was sitting around position 69
So the issue was not some mystical keyword puzzle. Google just was not getting much help from the page itself.
What I Changed On February 27
I did not touch the exercise logic. No URL changes. No big rewrite. That would have been drama for no reason.
Instead I added a reusable server wrapper, ExercisePageSEO, and fed it typed content from src/data/seo/*.ts. Each enriched page got:
- a visible H1 instead of the hidden one
- a short explanation block, around 150 to 200 words
- a “Quick Rule” box with the grammar formula and example sentences
- an FAQ section rendered as visible HTML
Quiz,BreadcrumbList, andFAQPageJSON-LD- related exercise links where they were missing
That pushed each page from almost-empty server HTML to something much closer to what a real educational page should look like.
How I Measured It
This part mattered because the whole site was already climbing fast.
If I had just compared before and after screenshots and started congratulating myself, that would have been a bit lazy. So I used the rest of the site as a control group.
- Before window: February 1 to February 26, 2026
- After window: February 27 to March 13, 2026
- Test group: 14 enriched pages
- Control: 416 non-enriched pages over the same dates
That let me ask a better question: did these pages improve faster than the general site trend?
What Happened By March 15
They did.
| Metric | Enriched Pages | Rest of Site |
|---|---|---|
| Average position change | -10.1 | -4.8 |
| Clicks per day growth | +163% | +85% |
| Impressions per day growth | +140% | +78% |
So the enriched pages were improving at roughly 2x the rate of the baseline.
A few page-level examples were even clearer:
zero-conditional-exercises: 42.6 -> 10.0modal-should: 45.7 -> 18.8third-conditional-exercises: 30.0 -> 8.5modal-must-have-to: 32.2 -> 18.2
That was enough for me. Not “SEO guru on a mountain” enough. Just enough to say the pattern was real and worth scaling.
The Annoying Bits
The experiment was useful partly because it showed what still needed fixing.
Two pages had messy readings because Google was indexing both the slash and no-slash versions:
articles-a-vs-anpassive-voice-present-simple
That split ranking signals and made the raw exports more annoying to read than they should have been. Which is a decent reminder that technical SEO can still mess up a perfectly sensible content test.
There was also one proper regression:
conditionals-with-modalsdropped from 20.9 to 33.5
That looked less like “content does not work” and more like search intent mismatch. I had improved the page structure, but the explanation and FAQ probably were not matching what people actually wanted when they searched that phrase.
That bit was useful too. Better structure helps, but it does not give you a free pass on relevance.
Why I Rolled It Out Everywhere
By March 20, 2026 I had enough evidence to stop treating it like a tiny experiment and start treating it like a product pattern.
So I rolled the same approach out across 834 exercise pages in 37 grammar modules.
The broader site trend at that point was already a bit mad:
- December 2025: 5.4 Google clicks/day
- January 2026: 30.4 clicks/day
- February 2026: 142.1 clicks/day
- March 2026 up to March 13: 278.6 clicks/day
The enrichment experiment obviously did not cause all of that on its own. But it was one of the first SEO changes I could point to and say, with a reasonably straight face, “yes, this one seems to be doing something beyond vibes.”
What I Took From It
The big lesson was boring in the best way.
Search engines want pages that explain themselves properly. Visible headings. Real copy. Examples. Structured data. Internal links. Enough server-rendered context that the page looks like a page and not a React loading state in a trench coat.
Fair enough, really.
Related: