01 · search that machines trust
Technical SEO & AI search
Google still matters. So do the answer engines quoting whoever they can parse. I make sites legible to both — crawlable, fast, structured, and cited — then prove it with instruments, not decks.
the discipline behind campaigns for Pitney Bowes and Urban Outfitters — and the engine this site grades itself with · scroll — the instrument is live ↓
the instrument — live crawl waterfall — this page auditing itself
abstracted from this site's real route table
crawl — this site's route map, drawing itself
bars abstract · status real
the work — manifest
w-01
Deep technical audits that end in shipped fixes, not backlogs — crawl architecture, rendering, indexation, Core Web Vitals to green.
w-02
AI-search readiness: structured data as a connected entity graph, llms.txt, clean machine-readable mirrors, citation-worthy page anatomy.
w-03
Migrations without casualties — URL inventories frozen, redirect maps generated and tested, rankings watched through the flip.
w-04
Ecommerce at catalogue scale: faceted-navigation control, canonical discipline, feed and category architecture that pulls the long tail into search.
w-05
Performance rescue: server response, caching layers, render paths — found at the root, fixed at the root.
w-06
Measurement that tells the truth: Search Console, server logs, and answer-engine visibility read together.
how it's different
I implement. The audit and the fix are the same engagement — no handoff to a dev team that never comes.
This site is the reference: run the Agent-Ready Grader at /lab against your site, then against this one.
Training-bot blocks are respected, citation bots are courted — the distinction most audits miss.
method — how it runs
01
read the machines' view
crawl the site the way each reader does — Googlebot, citation bots, agents — and reconcile what robots.txt promises against what the server actually serves.
02
find the root, not the symptom
rankings drop for reasons that live in server configs, render paths, and information architecture. The diagnosis names the mechanism, with evidence.
03
ship the fix personally
the same hands write the recommendation and the commit — templates, redirects, schema, server rules. Nothing waits in a backlog.
04
watch it hold
search console, logs, and vitals monitored through the change. If it moves the wrong way, it's caught in days, not quarters.
manifest — what lands on your desk
jm/technical·01
- 01
a prioritised technical audit that reads like an engineer wrote it
- 02
shipped fixes — commits, configs, schema, redirects
- 03
a frozen URL inventory + tested redirect map on migrations
- 04
an entity graph + llms.txt for the machine readers
- 05
a measurement baseline you can hold future work against
signed — one operator, no handoffs
05 items
engagement — the terms
Audits from a fixed fee. Retained technical ownership by the month. One person end to end.
proof — on the record
pitney bowes — named engagement
nameda full technical SEO campaign behind the franking-machine offering in the UK and US.
the file →urban outfitters — named engagement
nameda complex technical SEO campaign for the ecommerce store — architecture, crawl, and indexation.
the file →verified readout
verifieda UK tile retailer's server response taken from 2.2 s to 47 ms — plotted live on the work page.
watch it plot →asked straight — answered straight
01What does an AI-search audit actually cover?
Everything a machine reader touches: crawler access reconciled against what the server really serves (citation bots courted, training-bot blocks respected), content extractability, structured data as a connected entity graph rather than scattered snippets, snippet eligibility, llms.txt and machine-readable mirrors, and a share-of-answer baseline so visibility in assistants is measured, not guessed.
02Do you implement, or hand over a recommendations document?
I implement. The audit and the fix are the same engagement, in your codebase — WordPress estates, headless builds, custom stacks. Nineteen years of watching backlogged recommendations die quietly is why the practice works this way.
03A migration lost our traffic. Can it be recovered?
Usually, and the method is unglamorous: freeze an inventory of what existed, rebuild the redirect layer and test every hop at runtime, read the log files for where crawl budget actually went, then stage the recovery and watch it through. Migration autopsies are a specialism precisely because so few are planned as boringly as they should be.
04How do you measure visibility in AI answers?
From the ground up: citation and agent bots identified in server logs (this site publishes its own machine guestbook), answer sampling across the assistants that matter to you, and share-of-answer tracked over time. Instruments you keep — not a black-box dashboard subscription.