Problems
Start from what is actually happening.
Six situations
Which one sounds like you?
- Usually urgent
Organic traffic dropped
Something changed. There is always a date, and the date narrows the cause.
“Our organic traffic dropped and nobody can tell me why.”
First three things we check
- 1.Establish a clean baseline
- 2.Find the exact date
- 3.Check indexing and coverage
All 9 steps
- Time-sensitive
Traffic but no leads
Usually a traffic-quality problem, a conversion problem, or a measurement problem — and they need different fixes.
“We get traffic. It does not turn into anything.”
First three things we check
- 1.Check whether the leads exist
- 2.Classify the traffic by intent
- 3.Measure conversion by landing page
All 7 steps
- Usually urgent
Analytics cannot be trusted
When the numbers are wrong, every decision made on them is a guess with a chart attached.
“Our GA4 and CRM numbers do not match and I do not know which to believe.”
First three things we check
- 1.Inventory everything that collects data
- 2.Classify every event and conversion
- 3.Check what conversions actually measure
All 7 steps
- Time-sensitive
A website migration is coming
Migration losses are avoidable, but only with work that happens before launch — not after.
“We are replatforming and I am worried about losing our organic traffic.”
First three things we check
- 1.Build the URL inventory
- 2.Score every URL by value
- 3.Map old to new as rules
All 8 steps
- Structural
A large site with no clear system
Large sites need rules, thresholds and monitoring — not a longer list of tasks.
“We have thousands of pages and no idea which ones matter.”
First three things we check
- 1.Segment the site by page type
- 2.Measure indexation by type
- 3.Map crawl budget
All 7 steps
- Time-sensitive
AI search does not mention us
Measurable, improvable, and not guaranteeable. Anyone promising guaranteed AI placement is selling something else.
“When people ask AI about our category, our competitors come up and we do not.”
First three things we check
- 1.Build a fixed prompt set
- 2.Establish a baseline across surfaces
- 3.Identify which sources are being used instead of you
All 7 steps
Or use the diagnostic
Eight starting points, no form required.
You said
“Our organic traffic fell and nobody can explain it.”
What we would investigate
Establish a clean baseline
Pull the full available Search Console history and GA4, so seasonality and year-over-year comparison are possible before anything is concluded.
Find the exact date
Day-level analysis per template and per query group. Different sections often have different dates, and that pattern alone eliminates whole categories of cause.
Check indexing and coverage
Did pages leave the index, or did they stay indexed and lose position? These are completely different problems with completely different fixes.
Compare technical state before and after
Canonicals, robots, status codes, redirects, rendering and internal links around the date. Most drops that are called algorithmic turn out to be a release.
Compare content changes
Pages edited, consolidated, pruned or moved. Removing pages with real demand is a common self-inflicted cause.
Check whether the drop is real
Cross-check GSC against GA4 and the CRM. A consent or tracking change can produce a drop in reported traffic while actual traffic is unchanged.
Compare competitors and result layout
Did someone take those queries, or did the result page itself change shape so the same position now earns fewer clicks?
Check links to affected URLs
Lost backlinks or internal links to the specific pages that dropped — measured on the affected cohort, not site-wide.
Line up external events
Known search updates, seasonality, market events and your own campaign history against the isolated date.
You said
“We get visitors. They do not become anything.”
What we would investigate
Check whether the leads exist
Reconcile GA4 against the CRM first. A meaningful share of “we get no leads” turns out to be “we get leads and cannot see them”.
Classify the traffic by intent
Split queries into brand, commercial, informational and irrelevant. Informational traffic converting at a low rate is normal, not a failure.
Measure conversion by landing page
Conversion rate per template and per page, not site-wide. A site-wide average hides both the good pages and the broken ones.
Check the conversion path on each key template
Is there a clear next step? Is it the right next step for that buyer stage? “Learn more” everywhere is a real cause of this problem.
Test the forms for real
Submit them. Check validation, error handling, mobile behaviour, required fields, spam filtering and whether the submission actually reaches the CRM.
Analyse lead quality by source page
If leads exist but sales calls them poor, the problem is which searches you are winning — not the conversion rate.
Compare against paid performance
If paid converts on the same landing pages and organic does not, the difference is usually intent, not the page.
You said
“GA4 and our CRM disagree and I do not know which to trust.”
What we would investigate
Inventory everything that collects data
Every GA4 property and stream, every GTM tag, trigger and variable, every ads pixel, and every CRM field that is supposed to capture source.
Classify every event and conversion
Correct, wrong, duplicated or legacy. Most containers we open have tags nobody currently owns.
Check what conversions actually measure
A conversion defined on a thank-you page view is not the same as a confirmed submission — and behaves completely differently when routing changes.
Perform real actions
Submit a real form, complete a real purchase or signup, click a phone link, and trace each one through GTM, GA4, the ads platforms and the CRM.
Test both consent paths
Granted and denied. A consent implementation that drops data instead of modelling it can move reported traffic overnight for a compliance reason nobody connected to the chart.
Check cross-domain and subdomain behaviour
One journey becoming two sessions inflates traffic, destroys attribution and creates enormous “direct”.
Reconcile against the CRM
Match front-end conversions against CRM records, quantify the gap, and explain each contributor to it rather than averaging over it.
You said
“We are replatforming and I am worried about losing traffic.”
What we would investigate
Build the URL inventory
Crawl the old site, then add Search Console pages, GA4 landing pages, backlinked URLs and conversion pages. One list, every URL.
Score every URL by value
Traffic, conversions, revenue, backlinks and query value. This is what turns a redirect map from a guess into a protection plan.
Map old to new as rules
Rules wherever a pattern exists, explicit pairs where it does not. A rule can be tested against thousands of URLs; a spreadsheet cannot.
Specify what must carry across
Metadata, canonicals, structured data, internal links, sitemaps and robots — for both the staging and production environments.
Specify analytics continuity
Events, conversions, dataLayer and cross-domain behaviour on the new build, tested on staging before launch.
Test on staging
Crawl staging, run the redirect rules against the real scored URL set, verify rendering and markup, validate the tags.
Define launch-day pass/fail criteria
And rollback triggers. “Should we roll back?” should be answered by criteria agreed in advance, not by argument at 2am.
Monitor for 30 days
Day 1 critical QA, day 3 indexing and errors, day 7 traffic comparison, day 14 page-level comparison, day 30 retained-versus-lost report.
You said
“We have thousands of pages and no idea which ones matter.”
What we would investigate
Segment the site by page type
Not by URL. Every analysis after this — indexation, traffic, conversion, quality — is done per type, which is the only way large-site data becomes readable.
Measure indexation by type
What share of each page type is indexed, and what share of the indexed pages earn anything within 90 days.
Map crawl budget
Where crawl requests actually go, using server logs where available. Usually a large share goes to parameters, facets and duplicates.
Analyse the internal link graph
Link depth and internal links received by the pages that carry commercial value, versus the pages that receive links because of the navigation template.
Find the duplication and cannibalisation clusters
At scale these are systematic rather than incidental, which means they can be fixed with a rule instead of page by page.
Define a quality floor per page type
The minimum data or content a page must have to deserve indexation. Then measure how many pages currently clear it.
Design indexation and canonical rules
Which combinations are indexable, which are canonicalised, which are blocked, and how sitemaps are segmented so coverage is diagnosable.
You said
“AI assistants recommend competitors instead of us.”
What we would investigate
Build a fixed prompt set
Discovery, category, comparison, recommendation and brand questions, written down and frozen so results are comparable month to month.
Establish a baseline across surfaces
Run the set on the AI search surfaces available to us and record mention, recommendation, citation, source used, competitor mention and factual accuracy.
Identify which sources are being used instead of you
This is often the most actionable finding: a directory, a review site or a competitor blog is answering questions about your business.
Audit entity clarity on your own site
Can a machine extract who you are, what you do, who you serve, where, how you price and who works there — in plain sentences, without inference?
Check consistency off-site
Website, LinkedIn, directories and profiles should state the same core facts. Contradictions across sources make every source less usable.
Check technical access
Are the pages that explain your business crawlable and renderable without JavaScript? Are relevant crawlers blocked in robots.txt by accident?
Assess citable material
Original data, research or worked examples give these systems something to reference. Marketing copy gives them nothing.
You said
“The right people arrive and still do not convert.”
What we would investigate
Confirm the conversion data is real
Submit the forms ourselves and trace each one through GTM, GA4 and the CRM. A meaningful share of low-conversion cases are measurement problems.
Measure conversion by template, not site-wide
A site-wide rate hides both the pages that work and the pages that are broken. Almost every real conversion problem is concentrated.
Check whether the next step matches the buyer stage
“Learn more” on a page reached by a high-intent commercial search is a genuine cause of lost conversion — the CTA is the wrong commitment level.
Test the form and checkout on mobile
Required fields, validation messages, error handling, tap targets, and whether spam filtering is silently discarding real submissions.
Compare against paid on the same pages
If paid converts and organic does not on identical pages, the difference is intent — a traffic problem wearing a conversion costume.
Look at what the page does not answer
Scope, price guidance, process, what happens next, proof. Buyers who cannot answer these leave and shortlist someone who told them.
You said
“We keep publishing and nothing improves.”
What we would investigate
Check whether new pages are even indexed
Publishing into a site that is not indexing new URLs produces exactly this pattern, and it is invisible unless someone looks at coverage by page type.
Look for cannibalisation
Several pages competing for the same query keeps all of them mid-page. At scale this is systematic and fixable with a cluster model.
Compare demand against what you publish
Content briefed from volume rather than buyer stage produces informational traffic and no commercial movement. That is a structure problem, not an effort problem.
Audit internal linking
If links are generated by the navigation template rather than by page importance, the pages you care about never accumulate anything.
Check the technical floor
Rendering, canonicals, status codes and crawl distribution. Content work underperforms until the technical floor is solid, and it is cheaper to check than to assume.
Verify the measurement
Some stalled programmes are actually working and being measured badly — brand and non-brand averaged together is the most common version.
8 starting points. If none of them fits, tell us what is actually happening on the contact form — a real description is more useful to us than a category.