The situation we started from.
dongolabs.com had the usual signs of a technically competent site with no organic demand: canonical URLs, JSON-LD, breadcrumbs, robots directives, a generated sitemap and llms.txt files — and 24 clicks with 527 impressions over 90 days, roughly an 8-click monthly run rate. 186 pages were built, 140 were indexable, and the sitemap was padded with portfolio URLs that were themselves marked noindex. The temptation in that position is to publish 50 more pages and call it a content strategy.
Constraints the work had to respect
- No fabricated metrics, testimonials or client outcomes — including in our own case study.
- No mass-produced pages and no automated link acquisition.
- No mass-noindexing of useful pages just because they are short.
- Every claim had to be checkable in the repository, in the build output, or in Search Console.
Approach
Ask for a prioritisation, not an essay
We used Jev, a System One model that returns typed judgments and probabilities rather than prose, to rank candidate interventions. It put a major technical defect as the primary root cause at only 12% — useful, because it ruled out the comfortable explanation that the site was broken rather than unproven.
Commit to one batch and record the confidence
The selected batch — sitemap and indexability cleanup, contextual internal links, and SERP title fixes — scored 91% at 0.88 confidence. Post-implementation validation came back at 92% with an estimated 5% regression risk and 84% release readiness. Those numbers are recorded in the repository, including the ones that were unflattering.
Replace the word-count rule with an intent-aware policy
Instead of a universal minimum, each page class earns its own threshold and structural requirements: portfolio pages 180 words, tools and conversion pages 350, services, locations and industries 700, editorial 900 — each also checked for headings, answer structure and real internal links.
Keep execution deterministic
The model made bounded semantic decisions. Code did the work and the checking: sitemap generation, the link sections, the title fixes, and a content-quality checker that runs in warning mode locally and strict mode in CI.
What shipped
- Removed noindexed portfolio URLs from the sitemap, leaving 138 clean URLs and zero noindex entries.
- Added reusable contextual-link sections across tools and conversion pages, and corrected invalid service URLs.
- Shortened two overlong title tags that exceeded the SERP display budget.
- Added factual project context to featured portfolio pages and an intent-specific decision layer to industry pages.
- Added genuinely useful briefing content to the contact page and a 'Who this is for' section to the free-audit page.
- Shipped an intent-aware content-quality checker with a warning mode and a strict CI mode that fails the build.
Outcome, stated honestly
The site is now technically clean, structurally consistent, and instrumented well enough that the next 30 days will produce evidence instead of opinions. The model's own 30-day read was 'technical recovery and early visibility' at 97% — explicitly not a traffic spike.
What is verified
- Sitemap reduced to 138 URLs with zero noindexed entries.
- All 186 built pages pass the strict content-quality gate.
- Every prioritisation, validation and forecast judgment is recorded with its probability and confidence.
- Search Console indexing and query data was selected as the first measurement set at 100% probability.
What we are not claiming
- No traffic increase is claimed. Probability that impressions beat the baseline trend within 30 days: 49% — effectively a coin toss.
- No doubling of clicks is claimed. Probability from this implementation alone within 30 days: 12%.
- No guarantee is offered. Probability that any implementation could honestly promise a 90% chance of doubling clicks in 30 days: 3%.
- No authority or backlink outcome is claimed; that is a separate phase, to be judged against real referring domains.
What carries over to client work
- A calibrated probability is more useful than a confident recommendation: 12% for 'technical defect is the root cause' redirected the entire plan.
- Thin-content policy should be a function of page intent, not a single word count applied to a portfolio page and a pillar article alike.
- If a gate does not fail the build, it is a suggestion, not a standard.
- The honest version of a case study — including the forecast you refused to inflate — is the one a sceptical buyer can actually use.