Three domains, one SEO programme

North star: "seo agency" into the top 5, "ai seo agency" held in the top 5. Anti-metric: blog impressions. The blog carries 50.5% of site impressions at 0.09% click-through.

• Shipping is the constraint, not analysis. Several items here already have written specifications. If they are still unshipped at the next review, the click-through objective should be restated honestly rather than carried.
• Canberra and Hobart share an invisibility pattern. Diagnose once. If it is site-level, other location pages are affected and the fix is worth far more than two pages.
• Click-through is 0.15% against 6.56M impressions. The objective rests entirely on converting demand the site already earns, so watch click-through ahead of any volume metric.
• Australia is the content source for two other markets. Anything refreshed badly here propagates.

Page views and first visits are both flagged as conversions, so analytics reported 37,899 conversions against 129 real form submissions over 90 days.

True conversion rate is 1.61%, not 471%. Any UK performance claim made before 4 August 2026 was wrong by that factor. Until the events are corrected, report the UK on completed form submissions only, and never compare UK conversion totals to Australia.

The United Kingdom outranks Australia for enterprise and is level on ecommerce, while sitting 20 to 24 positions behind on the generic terms. That is not a weaker Australia, it is a differently shaped market. Defending and extending enterprise and ecommerce is a nearer win than dragging "seo agency" from 34.8.

The site earns 18 non-brand clicks per quarter from 59,189 non-brand impressions. There is also no analytics property, so the market is blind on sessions and conversions entirely.

The non-brand impressions are not spread randomly. They concentrate hard.

Roughly 33,000 impressions sit on ecommerce variants at positions 28 to 38. Meanwhile "seo agency" sits at 46.5 on 204 impressions, and "ai seo agency" has recorded one impression.

Converting a weak existing classification is far cheaper than building a new one. The US roadmap therefore leads on ecommerce, and states plainly that the north star here is a 2027 objective, not a 2026 one. Saying so is more useful than setting a target that will not be met.

Active work

• US1 Add structured data. Cheapest fix on the roadmap and a prerequisite for entity recognition.
• US2 Create an analytics property. Blocked on Lawrence.
• US3 Add case studies. 18 to 30.
• US4 Work the ecommerce cluster. The cheapest real ranking opportunity in this market.
• US5 Service page link pass. 14 pages.
• US6 Industry pages. Prioritise verticals already showing impressions: healthcare, contractor, ecommerce.
• US7 City and suburb expansion. Atlanta, Dallas and Austin pages exist as a base.
• US8 Receive and regionalise refreshed content. Highest-volume lever available, given the corpus is only 10 posts.
• US9 Google Business Profile. Atlanta is the known office.
• US10 Brand-confusion check. Confusion terms and client brand leakage appear in the ranked set. Treat the 2,761 query count as soft until resolved.

The markup itself is clean. Across every page checked: zero one-way annotations, zero targets that fail to resolve, zero missing self-references, zero conflicts with the page's own canonical. Where language targeting exists, it is textbook.

Counting only pages that genuinely have a counterpart in another market, because those are the only ones that need it:

The gap is about 13 real pages, and it is almost entirely the blog: the blog hub on all three domains, five or six article pairs that exist across Australia and the UK, plus the awards page on Australia and the US. The three llms.txt files also show as gaps but are plain-text spec files and cannot carry link tags, so they are not defects.

Why those pages were never linked

The cross-market articles are properly regionalised, not duplicated. "How do search engines work" runs 4,356 words in Australia and 3,532 in the UK. "How to do keyword research" runs 6,475 against 4,594, and the UK version is written explicitly for UK businesses. That is the right way to do it.

What is missing is only the annotation telling Google they are variants of each other. And in at least one case the slugs have drifted apart, so the pair cannot be matched by path at all: the UK version of the search-engines article canonicalises to a longer slug than the Australian one.

The market we are asking to compete has no machine-readable identity. It explains part of why 95% of its clicks are brand-only.

Verified live on 5 August 2026. It is inconsistent in both directions, so neither convention is being applied consistently.

A suspected tracking-parameter indexing issue is resolved. The tagged homepage returns 200 and canonicalises correctly. It also earns 400 clicks at 1.78% click-through, the best of any URL on the site, against the homepage's 0.51%.

Older findings carried into this roadmap were re-tested rather than inherited. One was resolved and struck off, one was still open and had degraded in an unexpected way. Neither status was assumed.

They rank number 1 in the UK for "seo agency", a 27,100 per month term. Verified on a live SERP, 11 August 2026, and corroborated by their own Semrush export. Their second URL sits at 47 for the same term.

What the win is actually built on

The uncomfortable half, and why it still matters

That number 1 position produces 85% of their entire estimated traffic from a single keyword. Meanwhile 78% of their 4,400 keywords sit at position 51 to 100 earning nothing, and their quality node ratio is 20.3% against a 50% threshold.

Read correctly, that is not a failure. It is extreme concentration. They built breadth to support the entity and put everything into winning one term, and it worked. The lesson is not that breadth pays. It is that the head term pays, and everything else is scaffolding for it.

On speed: their export shows 99.1% of positions unchanged and only 29 new keywords, so it cannot evidence how fast they got there. The snapshot shows a stable number 1, not a climb.

What we would take from it

• A minimum page-depth standard. They hold 1,000 words with no exceptions. We have no floor at all.
• The hub build. Proof and the child index live on the hub page, rather than being linked away from it.
• A published entity map, linked sitewide.
• The industry taxonomy question: under the service, or parallel to it. Free to decide now, a migration later.
• Not their breadth. 3,433 keywords earning nothing is the part to leave.

This is not a ranking example and should not be used as one. They hold 55 ranked keywords in Australia and are position 1 only for their own brand name. Everything else sits between 17 and 47 (DataForSEO, AU, 11 August 2026).

What is worth studying is the offer architecture on a single 11,723-word homepage:

• A named system. "Four Parts. One Chain." The offer has a name and a shape, rather than being a list of services.
• The problem stated before the solution. "You're Ready to Grow, But Disconnected Parts Are Holding You Back", then "We Find Exactly What's Costing You Leads".
• An explicit ICP, in the customer's own words. "Built For Hard-Working Aussie Business Owners". No ambiguity about who it is for.
• Proof as a headline number. "Helped Aussie Businesses Add Over $10m In Revenue".
• Competitor framing handled directly. "Why Other Marketing Options Fail", then "Why Our Approach Works".

This is the conversion layer, and it connects to the parked ICP positioning work rather than to the topical map. Their schema stack is Organization, WebSite, WebPage, BreadcrumbList and SearchAction, with no LocalBusiness markup despite the local positioning.

• Claude SEO cluster, 4 drafts. Each with a visual and structured data, all questions answered inline for a clean handoff.
• How to rank your social content on Google. Hub drafted with real channel data. Five platform spokes queued.
• Content backlog, 55 finalised drafts. 91 triaged to 50 topics plus 5 refreshes, all publish-ready. Cadence 2 to 3 per week.

• Awards. Global Search Awards in prep; SEL 2026 and APAC 2027 slates tracked separately.
• Employee branding programme. Target 100 videos by 31 December. Regional owners Sian, Anthony and Sophie.
• Personal brands, five people. Top action is deploying the consultant site.
• Lawrence video slate. Check the catalogue before new slate work, duplicates exist.
• Global talent and remote-first hiring. Needs a one-page scope brief. Unblocks operations and recruitment.
• Weekly deploy-day process. Draft with open questions, to discuss on Harry's return.
• ICP positioning experiment. Parked Needs board and marketing approval before any signal is edited.

• Digital PR link building by vertical. Pillar plus vertical spokes on whitepaper data: 4,368 placements, 1,027 journalist contacts, 1,982 cost rows.
• Local and business profile mini-cluster. Pairs with the profile workstream.
• Content calendar backlog, 42 items. 13 refreshes plus 27 new ideas. Refreshes first, they already rank.
• Five drafted, unpublished blog posts on AI search topics.
• Lead magnets: 6 SEO templates, 4 tools, 2 whitepapers, 7 content templates, 3 digital PR templates. All still to build.
• Negative SEO protection. Research and implement defences.
• Suburb page programme. 67 opportunities identified. Sydney cluster first, biggest anchors Parramatta and Penrith.

For prioritisation: Search Console first, then the testing platform, then third-party rank tools. For keyword discovery, breadth and SERP features: the third-party tools.

This has already changed decisions. A vendor reported one service page at position 92 when Search Console showed it in the top 10, and reported rankings for Canberra and Hobart that Search Console shows earning zero impressions. Acting on the vendor number alone would have produced the wrong work in both directions.

Monthly, and at the end of each quarter:

• Run the cross-source sweep. Read the synthesis, not the per-source dumps.
• Compare click-through and position deltas on every refreshed page.
• Run a technical crawl and confirm no broken internal links after framework changes.
• Confirm a test annotation exists per refreshed page.
• Sample five published drafts against the linking framework and check for double-linked keywords.