Research

SEO experiment register

This register holds 11 dated SEO tests on my own sites. 8 are still running, 2 are inconclusive and 1 failed. None has a clean win yet.

I never log client sites here. Each entry states what I changed, what I compared it with and how long I waited. Tests I wrote up after the fact are labelled that way. Tests I set up in advance are labelled as registered before the result.

Register opened 2 November 2026. Last updated 2 November 2026.

Running
8
Concluded
0
Inconclusive
2
Failed
1
Readback dates, 2 November to 8 December 2026
  1. EXP-004
  2. EXP-006
  3. EXP-008
  4. EXP-009
  5. EXP-010
  6. EXP-011
  7. EXP-007
  8. EXP-005

Left edge is 2 November. Right edge is 8 December. Each diamond marks one readback.

The entries

Newest change last. A running entry shows its next readback date and no outcome.

EXP-001Failedexploratory (logged after the fact)

Bulk title rewrite on three sites

Site
Three of my own sites. Two were large programmatic sites of about 144,000 and 776 pages.
Hypothesis
I wanted a click-through lift from new titles and meta descriptions. I did not write a hypothesis down before the change.
Change made
Title and meta description templates were rewritten on all three sites within about six minutes. ()
Comparison
None. No control pages.
Metric
Search Console average position.
Window
The days after the change.
Outcome
Positions I had logged in the 7 to 60 range moved to 50 to 80 or worse on all three sites. A Google core update was running at the time. I cannot separate the title change from the update. I now cap bulk rewrites at 20 pages per site per session. These figures come from my notes. I have not re-checked them in Search Console.
Log
  1. : Logged in this register after the fact. No readback is pending.
EXP-002Inconclusiveexploratory (logged after the fact)

Query-matched titles on four pages

Site
sunnypatel.co.uk
Hypothesis
A title that matches the real top queries of a page lifts its click-through rate.
Change made
New titles and meta descriptions on four pages. The local agencies guide got affordability wording. The GEO agencies guide got AEO wording. The UK companies guide got a freshness signal. The technical audit page got word order matched to its top query. I retitled the UK companies guide again on 6 October. ()
Comparison
The same pages before and after. No control.
Metric
Clicks and click-through rate per page in Search Console.
Window
Planned for 4 to 18 October 2026, which is 14 to 28 days after the change.
Outcome
I did not record a readback for the planned window. A later check compared two other windows: 2 to 29 September and 30 September to 27 October. GEO guide clicks fell from 7 to 3 between them. Whole-site clicks fell from 106 to 77. The first window contains the change date. This test cannot show whether the new titles helped or hurt. All figures are Search Console Web search for the whole domain, worldwide.
Log
  1. : Logged in this register after the fact. No readback is pending.
Evidence
The GEO agencies guide
EXP-003Inconclusiveexploratory (logged after the fact)

Direct-answer passage rewrites

Site
sunnypatel.co.uk
Hypothesis
Opening key passages with a direct answer makes pages easier for AI assistants to cite. Fixing weak proof claims should help in the same way.
Change made
Reviewed edits to ten pages, 35 site files in all. Examples: a plainer homepage introduction, a disclosed comparison method on the GEO guide, and new sections on the prompts tool that say which template to choose. ()
Baseline
Two matched windows of 27 days: 3 to 29 September against 7 August to 2 September. The source applies no country filter, so I treat these figures as worldwide.
Comparison
Matched windows before the change. No control pages. No window after the change was analysed.
Metric
GA4 sessions from AI assistants, Google AI impressions and Bing AI citations.
Window
Before the change only.
Outcome
All three measures were higher in the later window. Sessions were 41 against 28. Impressions were 4,428 against 3,308. Citations were 2,732 against 2,273. That rise came before the edits, so the edits cannot explain it. I have no post-change comparison and other site changes overlapped. No uplift is established.
Log
  1. : Logged in this register after the fact. No readback is pending.
EXP-004Runningexploratory (logged after the fact)

AEO buying-fit pilot

Site
sunnypatel.co.uk
Hypothesis
I expect the practice to fit consultant and audit requests better in AI answers. I expect clearer independent delivery, prices, ownership and supported evidence to coincide with that better fit. Removing unsupported proof and absolute claims should also cut wrong descriptions of the business.
Change made
Nine files across three pages, shared service data and public text summaries. Scope, fees and owners made explicit. Unverified proof cards, testimonials and citation deadlines removed. Claims about crawler access and schema corrected. ()
Baseline
47 sessions from eight AI assistant sources, 27 of them engaged, from 5 September to 2 October 2026.
Comparison
The same pages before and after, plus six fixed buyer prompts. Two prompts are held out until day 28. No control. The package changed together, so no single edit can take the credit.
Metric
GA4 sessions from the eight AI sources, and captured answers to the six prompts.
Window
Reviews at day 7, 14 and 28. Completed-window analytics at day 32.
Next readback
Outcome
No result yet. My first capture of ChatGPT answers stalled, so I have no baseline answers. A missing capture is not a zero. The protocol allows a descriptive comparison only.
Log
  1. : Logged in this register as exploratory. The protocol erratum says not to claim verified pre-registration.
Note
The frozen protocol file did not match its recorded hash. I kept the original and logged an erratum (v1.3). I do not claim verified pre-registration, so this entry is exploratory.
EXP-005Runningregistered before the result

Specific redirects for old recipe URLs

Site
shecookssheeats.co.uk
Hypothesis
Sending old URLs to their matching pages, instead of a generic hub, recovers search clicks.
Change made
A redirect map in the edge function. 90 of the top 150 URLs that went to a hub now go to a matching page, which covers 3,827 peak clicks. Nine redirects that led to a 404 were fixed and seven entries repointed. All 79 distinct targets returned 200 before the merge. It went live on 24 October 2026. ()
Baseline
932 URLs from the peak window, worth 12,458 clicks or 83% of peak clicks, redirected to a generic hub.
Comparison
The same site before and after. No control.
Metric
Search Console clicks per day for the whole site.
Window
45 days, to 8 December 2026. My threshold: more than 15 clicks a day by then, or I stop work on the site.
Next readback
Outcome
No result yet. A traffic fall in November 2025 came before the migration. I do not claim redirects will recover it.
Log
  1. : Registered with the change, before any result. Merged and live the same day.
EXP-006Runningregistered before the result

Berkshire hub de-target

Site
sunnypatel.co.uk
Hypothesis
Taking town names out of the Berkshire hub lets each town page own its own queries.
Change made
Town lists removed from the hub description, subtitle and intro. Town links point to the owning page. A compact enquiry form added to four town pages. Hub title and H1 unchanged. ()
Baseline
4 to 24 October, non-branded. Each figure is a share of impressions. Bracknell (530 impressions): town page 42%, hub 40%. Wokingham (335): town page 41%, hub 58%. Maidenhead (178): town page 52%, hub 46%. All these groups together earned 1 click.
Comparison
The same query groups before and after. No control. The merge of a small Reading local-SEO site on 27 October overlaps this window.
Metric
Share of impressions and average position per page for each town group. Submissions of the new town-page forms in GA4.
Window
9 to 23 November 2026, which is 14 to 28 days after release.
Next readback
Outcome
No result yet. I count it a success if the town pages hold the majority share of impressions and their positions improve.
Log
  1. : Registered with the change, before any result.
Evidence
The Berkshire hub
EXP-007Runningregistered before the result

Price-led title on one network page

Site
chargefinder.uk, a UK electric vehicle charging price site
Hypothesis
A title that leads with price lifts click-through rate on price queries.
Change made
A new title on one operator page. ()
Baseline
342 impressions at average position 5.09 and 0 clicks.
Comparison
Two operator pages with unchanged titles act as controls. They are not matched on traffic, so this is a rough comparison.
Metric
Query-level click-through rate, changed page against controls.
Window
A 14-day lag, then a readback around 16 November 2026.
Next readback
Outcome
No result yet.
Log
  1. : Registered with the change, before any result.
EXP-008Runningregistered before the result

301 merge of a small Reading site

Site
A small Reading local-SEO site, merged into sunnypatel.co.uk
Hypothesis
Redirecting the old site into matching sunnypatel.co.uk pages costs no Reading visibility. I expect a small gain at most.
Change made
Every checked old URL (22 in all) now returns one 301 to a live sunnypatel.co.uk page. No content moved. A change of address was started in Search Console. ()
Baseline
The old site had 6 clicks in 90 days at average positions of 40 to 69.
Comparison
The same Reading queries before and after. No control. The hub change of 26 October overlaps, and the old areas page now redirects to the Berkshire hub.
Metric
Impressions, clicks and position for Reading queries on sunnypatel.co.uk.
Window
9 to 23 November 2026.
Next readback
Outcome
No result yet. I expect cleanup, not a ranking change. The overlap means a movement cannot be pinned on the merge.
Log
  1. : Registered with the change, before any result.
Evidence
The Reading page
EXP-009Runningregistered before the result

Evidence repair on a provider guide and a schema fix

Site
sunnypatel.co.uk
Hypothesis
Direct case-study links, split source and update dates, and a description that no longer over-promises keep the guide's UK exposure from sliding. The schema fix is a correctness repair, so I only track its position.
Change made
Guide description cut back, byline dates split into sources checked and content updated, and 13 provider profiles linked to a case study on the provider's own site. The schema generator now emits a rating only when both value and count are present. Title, H1 and URL unchanged. ()
Baseline
Guide: 80 impressions and 32 visible queries in the week of 21 to 27 October. Schema generator, UK desktop: position 12.6 for schema markup generator and 12.5 for schema generator.
Comparison
The same pages before and after. UK and worldwide data are kept apart. No control, and the guide was already losing exposure before the change.
Metric
UK impressions and visible query count for the guide. UK desktop position for the two schema queries.
Window
13 to 27 November 2026, after recrawl.
Next readback
Outcome
No result yet. I do not claim this change caused the earlier decline.
Log
  1. : Registered with the change, before any result.
Evidence
The UK companies guide
EXP-010Runningregistered before the result

Audit page retitle and niche service page offers

Site
sunnypatel.co.uk
Hypothesis
A clear price in the title and a first-screen offer on each niche service page lift visibility and enquiries for audit and niche queries.
Change made
An old audit URL now 301s to the technical audit page, which has a new title with the fixed fee. Six niche service pages got first-screen offers, verified migration proof and buyer FAQs. Unsupported claims were removed. ()
Baseline
The query technical seo audit service had 1,954 impressions at position 26.0 from August to October 2026. The page was already slipping before this change.
Comparison
The same pages before and after. No control. Counts are small, so I will report raw numbers.
Metric
Impressions and position for technical seo audit service and seo audit consultant. AI-assistant landings on the six niche pages. Enquiries in my ledger.
Window
14 to 28 November 2026.
Next readback
Outcome
No result yet.
Log
  1. : Registered with the change, before any result.
Evidence
The technical SEO audit page
EXP-011Runningregistered before the result

Honest titles and DfE figures on 20 school pages

Site
catchment.school, a UK school admissions data site
Hypothesis
An honest title that says admissions should lift click-through and engagement. The school's own DfE first-preference and offer figures under the H1 should add to that. I expect no Bing loss.
Change made
Title, meta description and a figures block rewritten on 20 school pages from each school's own DfE row. All other school pages keep their titles. The same release added a Googlebot-only noindex to independent school pages, which are in neither group. ()
Baseline
A 28-day Search Console and Bing pull for all 40 URLs, due 3 November 2026. The release went live on 1 November. The pull itself is taken on 3 November, after release, and covers the 28 days before it.
Comparison
20 matched control pages on the same phase, matched on offers and share offered. The 21 best Bing pages are in neither group. I compare the change in each group, not the raw levels.
Metric
Bing clicks and click-through rate per URL. Google impressions and position. Groups with under about 30 clicks show raw counts and no percentages.
Window
Day 14 (15 November) and day 28 (29 November). I roll back early if cohort Bing clicks fall more than 30% against control at day 14 with at least 30 baseline clicks.
Next readback
Outcome
No result yet.
Log
  1. : Registered with the release, before any result.

How I run tests

I write the test down before the result, wait a fixed window and refuse to read small counts as trends.

Pre-registration

Before a result exists, I note the hypothesis, the baseline, the metric and the readback date. The dates on this page come from my own working notes. No outside service timestamps them. If I write a test up later, it carries the exploratory label and I treat it as a lead, not a finding.

Minimum window

I record a day-7 check but do not read results from it. Results come from a window of at least 14 days. My standard window runs 14 to 28 days. I keep UK and worldwide data apart and I use full days only.

Small-n rule

Under about 30 clicks in a group, I publish raw counts and no percentages. A result of not enough data is a valid outcome. I do not claim a cause when other changes overlap, and I say so in the entry.

Statuses

Running means the window is open. Concluded means the readback is done and the result is clear. Inconclusive means the data cannot separate the change from other causes. Failed means the metric went the wrong way, whether or not I can isolate the cause.

How to cite an entry

Cite the entry id, the page date and the status you saw. Running entries change when the readback lands.

Patel, S. (2026). SEO experiment register, EXP-005 (status: running). sunnypatel.co.uk. https://sunnypatel.co.uk/research/experiments/#exp-005. Register version dated 2 November 2026.

Every entry has its own anchor, such as #exp-001. Describe an exploratory entry as exploratory.

Want a test like these on your site?

The free 20-minute diagnosis is the place to start. I look at your site and tell you what is worth testing first.

Book the free diagnosis