A managing director sends me a 180-page PDF pulled out of an off-the-shelf tool. Three hundred and forty errors, forty warnings, a score of 62 out of 100. His question fits on one line: where do we start?
The answer is nowhere in the document. A tool can tell you what is broken. It cannot tell you what it is costing you.
An audit that does not rank its findings is not an audit, it is an inventory. The machine already does inventories on its own.
Here is the order in which I go through a site, and above all why that order. I keep to it on every project, from a thirty-page company site to a catalogue with forty thousand references.
1. Are the pages that make money actually indexed?
I do not open a crawler before I have opened Search Console. A page missing from the index is worth zero, however beautiful its tags are. Everything else is polishing thin air.
Page indexing report first, then the URL inspection tool on the handful of pages that matter. Two points that Google's documentation states plainly and that I regularly see misread. First, the "URL is on Google" status means the page is eligible to appear, not that it does: the tool does not check every condition required to show up. Second, what it displays comes from the last indexed version, not from the live page. If a fix has just gone out, you need the live test to see it.
I once found a noindex tag left by a developer on an entire product page template after a redesign. The client's tool did flag it. Line 112 out of 340, between an image with no alt attribute and an outbound nofollow link, at the same displayed severity.
2. Where does the money come from today?
I ask for the list of pages that carry revenue before I look at anything technical. Without it every fix looks equally worthwhile, and that is how teams spend weeks on pages that will never see a customer.
In practice: Search Console pages sorted by clicks, cross-checked against orders or quote requests actually recorded. On most of the sites I take over, a small handful of URLs carry the bulk of the demand. That is where the audit is decided, and nowhere else on the first pass.
That cross-check also settles what comes next. A technical error on a page that earns nothing stays a technical error. On the page that brings in a quarter of the enquiries, it becomes a priority. Same defect, two different treatments. The shopkeeper's view decides, not the technician's.
3. Does Google see the same page I do?
Right-click and "view source" proves nothing on a modern site. What counts is the rendered HTML, the one Google gets after running the JavaScript. URL inspection gives it to you: a screenshot of the page as it was rendered, the resulting code, the loaded resources, the JavaScript output.
When there is a gap, it is obvious. On one catalogue I watched product descriptions arrive through an external call fired after load: perfect in the browser, absent from what Google rendered. The site did not have a content problem, it had a content delivery problem. No tool report would have phrased it that way for me.
Same screen, one field I always read: the canonical Google selected, which only shows on indexed data. Which brings me to the next question.
4. How many versions of the same page exist?
Facets, sort orders, tracking parameters, regional variants, http and https living side by side: a five-year-old site almost always has several addresses for the same content. Google says so without any drama, some duplicate content is normal and is not a violation of its spam policies. It clusters the pages it considers equivalent and picks one, the canonical, which it will crawl more regularly than the rest.
The line worth keeping from that documentation is elsewhere: indicating a canonical preference is a hint, not a rule. You suggest, Google decides. Hence the only check that matters: compare, in URL inspection, the canonical you declare with the one Google kept. When the two diverge on a page that makes money, you have a real subject, and often the cause of a drop nobody could explain.
5. Did the site fall, or did demand?
This is what I ask before accepting a "we lost Google" audit. More than once I have watched that panic dissolve into a curve doing exactly what it did the year before, give or take a fortnight.
Google describes the method and it is a good one. Run the performance report over sixteen months to expose seasonality. Compare the period with the previous one. Then check the queries that lost the most clicks in Google Trends, to find out whether the drop hit the whole web or only you.
A simple reading already points the way. Impressions and clicks falling together: the cause sits upstream, technical, algorithmic or seasonal. Impressions flat and clicks collapsing: your title and snippet no longer earn the click, or a competitor is showing a richer result than yours. Two opposite diagnoses, two plans of attack with nothing in common.
Google adds two pieces of advice I take as my own. Do not fixate on absolute position, since impressions and clicks are the real measure. And avoid radical changes on a page that is already performing. That second one has saved more traffic than most of the optimisations I have shipped.
What I do with the tool's report
I do not bin it, it catches things the eye misses. I simply reread it with one extra column in mind: what does this actually cost at this client.
| What the tool flags | What it really costs | What I do with it |
|---|---|---|
| Duplicate title tag on 240 addresses | Almost always one template or facet defect, not 240 separate problems | I go back to the template and fix it once |
| Page set to noindex | A complete blocker on that page | Checked immediately on the pages that carry demand |
| Missing meta description | Nothing on ranking, a great deal on click-through | Handled as a batch, never at the top of the plan |
| Several H1s, irregular heading order | Marginal, unless the structure stops you understanding the page | I note it, I do not make a project out of it |
| Internal links pointing through a redirect | Real on a catalogue with tens of thousands of addresses, trivial on fifty pages | Priority indexed on the size of the site |
| "SEO score" of 62 out of 100 | No decision value at all | It does not go into my reports |
On that last row I am not alone. Google published a whole page on third-party tools and it leaves no room for doubt. Those tools have no access to its internal ranking data. They cannot guarantee performance, and their predictions are their own. Google also states that it does not evaluate these services and that using a tool guarantees no ranking success. A score out of 100 is therefore a vendor's opinion. An opinion can be useful; it does not set your order of priorities, your revenue does.
The most expensive mistake
Fixing things in the order of the report. That order comes from the vendor's severity grid, applied the same way to a brochure site and to a marketplace. It knows nothing about your margins, your converting pages, or the redesign scheduled in three months that will make half the fixes pointless. The first sort of an audit is done with the director and the profit and loss account open, not with a spreadsheet sorted by colour.
What I hand over in the end
Not three hundred and forty lines. Three named workstreams, a dozen precise fixes each attached to a page or a template, and the order in which to take them. For each fix, the scope in volume: twelve titles to rewrite, one template to rebuild, three redirects to put in place. It is the developer or the in-house team who says what that demands of them, not me.
And one thing tools never produce: the list of what we are not doing this quarter, put in writing. An audit that gives nothing up has prioritised nothing. It has merely moved the problem to the client's side of the table.
These five questions do not replace the full technical pass, they make it useful. If you want the exhaustive run-through, my 47-point SEO audit checklist and the article on the four phases of an audit cover the detail. Measuring the impact is prepared upstream, on the analytics and tracking side; and on WordPress, a good share of what tools flag comes from the page builder and the plugins rather than from the content.
In the crucible: what is left once the report has burned
A tool export is ore. There is gold in it, mixed with a lot of rock. The whole craft consists of heating it until only the three or four fixes that move a curve remain, then placing them in the right order. The rest will wait, and it will often wait a very long time without anyone noticing.
That is also why I always start with the same questions. They cost nothing to ask, and they stop you paying for polish on pages Google never indexed.
Sitting on an audit report nobody knows how to use?
Send it over with the pages that make your revenue. I will hand back the order to take them in and what we drop. Worst case you confirm what you already knew, best case you save a quarter's work on pages that earn nothing.