Changelog

This is what we ship on our own site, dated and specific. We do not ask a client to trust a process we are not running on ourselves, so this is the record. The log starts on 26 August 2026.

26 August 2026

Area pages

• Rewrote titles, meta descriptions and focus keywords on 194 city pages. Each title shape was derived from that page’s own Search Console queries rather than dropped in from one template. Quality gates enforced before writing: zero titles over 60 characters, zero duplicates, zero descriptions outside 115 to 155 characters.
• Found 143 city pages carrying Rank Math FAQ blocks, 1,939 written questions in total, that had never emitted a single line of schema. The oldest dates to February 2025, so that content had been sitting unused for around 18 months. All 242 area pages now emit FAQPage and Service schema, verified on the front end.
• Repaired 141 pages whose FAQ blocks had been saved in an invalid state, and cleaned escaped markup out of 1,529 question titles.
• Recovered two orphaned pages that held complete local data but rendered no template at all, because the template attached by author and theirs was wrong. Dayton, Ohio went from a bare page to 3,116 words of data that had already been entered and never displayed.

Blog

• Published 8 new technical articles on the WordPress database and site maintenance, rewrote one from scratch, and refreshed 9 older posts for AI Overview eligibility: FAQ schema, comparison tables and primary-source citations added to each.
• Removed rel="nofollow" from editorial citations across the blog. It is meant for advertising and user-generated content, not for citing WordPress core documentation, and it was quietly withholding a quality signal.
• Rebuilt the blog image pipeline. Featured and in-body images are now rendered from HTML and CSS straight to WebP at quality 84, roughly 20KB each against 150KB for the PNGs they replaced.
• Scheduled two further articles for 1 and 8 September.

Measurement

• SERP-tested 53 candidate queries across two passes in a live browser at 1440 by 778, measuring the pixel depth of the first organic result on each one. The finding: verbatim identifiers (table names, error strings, config constants) put the first organic result at 196 to 618 pixels, above the fold, while natural-language questions can push it past 2,000 and out of reach whatever position you hold. We now measure every query before writing anything, and we skip the ones that fail. Our first attempt at this rule blamed the People Also Ask block; a second pass disproved that, so the rule is now depth measured directly rather than inferred from what else is on the page.
• Disproved our own working assumption. We had concluded that AI Overviews were absorbing the clicks our articles used to get. They are not. Our best-performing post has an AI Overview sitting above it and holds a 15.6% click-through rate. Pixel depth is the variable that matters; the Overview is not.