Measuring and steering:the feedback loop

A report runs straight, a loop goes round

Most SEO reports are a dead end street. They show figures, someone nods and next week the same report comes with different figures. The difference between a report and a loop lies in one step: it gets checked whether what was proposed last week has also been carried out and what it did.

Only through that does learning arise. Without that step you gather data and you still decide on feeling. If the fifth step is missing, it is not a circle but a straight line.

Search Console first, because that is where the figures come from

Google Search Console is a free service that shows on which searches your site appears, how often, how often it gets clicked and at which average position. It is the only place where you see what happens before someone comes to your site. An analytics package shows what visitors do afterwards. You need both and Search Console first.

Setting it up takes twenty minutes if you can get at your DNS.

1. Sign in at search.google.com/search-console with a Google account of the company and not of an employee who leaves tomorrow.
2. Add your site as a property. Choose a domain property if you can set a DNS record, because that catches all subdomains and both protocols in one go. If that does not work, choose a URL prefix.
3. Confirm the ownership. A domain property goes through DNS. A URL prefix can also go through an HTML file, a meta tag or a link with your analytics package.
4. Submit your sitemap. That is set out step by step in chapter 1.
5. Wait. The first days the report stays empty. After a week or two you have something to look at.

Two practical points. Search Console keeps about sixteen months of history, so export monthly if you want to compare over a longer period. One property is enough for a multilingual site: you filter per language folder instead of managing four properties.

The five steps

1. Measure. Search data from Search Console, behaviour from your analytics package and a technical check of your own of your most important pages. Every measurement you keep as a snapshot, because without history there is nothing to compare with.
2. Explain. Figures alone say nothing. It is about the why: which page moved and which change went seven to fourteen days before it. For that you keep an event log in which every adjustment stands with its date.
3. Propose. At most three actions a week, each with a reason. Three carried out changes beat ten noted ideas. If there is nothing to do, that is written down too.
4. Change. The execution stays with a human. Automatic publishing takes away the control at exactly the place where quality arises. An approval step costs minutes and prevents weeks of repair work.
5. Check. Seven days later you lay the figures of the page concerned next to those from before. You note the result honestly, also when there is no result. Actions that stay open for three weeks lapse without blame.

What you measure, in three layers

The third layer is the only one you have to build or have built yourself. A script of about two hundred lines fetches your key pages and reports only the deviations from the thresholds in chapters 2, 3, 4 and 7. If it reports everything, nobody reads it any more after three weeks.

Two mistakes come back often here. The first is checking only the homepage: a telephone number on the contact page exists even when it is missing on the homepage. The second is adjusting the measuring instrument as soon as your own site scores badly. Anyone who changes a key figure because the outcome disappoints measures nothing any more.

The action list and its four statuses

Every proposal goes into one file with a date, the page it is about, the reason and a status. Four statuses are enough.

Lapsed is not a failure. A list on which everything stays gets ignored within two months. The most important status is measured, because almost everybody skips it. An action that was carried out and never checked teaches you nothing.

Keyword research without a subscription

Paid tools show estimated volumes. Free sources show real behaviour. For an SMB the second is usually more usable.

Your best source is your own Search Console. Filter in the performance report on searches with many impressions and few clicks: those are subjects for which Google already shows you while your page does not answer the question. The second list is the searches where you stand at position eight to twenty. There the gain lies within reach of one better page.

Google Autocomplete is the second source. What the input field suggests is what people really type, per language and per country. The keywords under each chapter of this guide come from there, mined on 3 August 2026 for the Netherlands and Belgium.

Google Trends gives the direction over twelve months and per region. Absolute numbers you do not get from it. For the question whether a subject grows or shrinks that is enough.

None of these sources tells you what a subject is worth. You only know that by writing one page about it and looking at what it does.

Four rules that keep the loop honest

Signal strength comes before the urge to act. At a few dozen clicks a week the fluctuation from week to week is almost entirely noise. Anyone who reacts to that steers on chance. Conclusions about content you draw over several weeks. Technical faults you do repair straight away.

Only act at full certainty. Every action has an unambiguous reason. Everything that is doubtful gets reported and stays put. That holds especially for advertising: you only exclude a search term when it cannot possibly come from a customer.

Follow up actions. Every round starts at the action list and not at the new figures. First close off last week, only then look at what came in. This is the step that almost everybody leaves out.

Check the loop itself monthly. Once a month ask two questions. Which part of the report never led to an act? That may go. Which blind spot came back repeatedly? That belongs in it. A measuring system that never gets simplified grows over and gets ignored.

The monthly competitor block

Once a month and not weekly, because competitors do not change their approach in a weekly rhythm. Four sources, all four free, together about twenty minutes.

• Google Ads Transparency Center. Shows every active search advertisement of a company including running time and landing page. You see which promise your competitor puts in the headline and where his budget goes.
• Meta Ad Library. The same for Facebook and Instagram, publicly consultable for every advertising account.
• Google Trends. Compares the demand development of your core terms over twelve months and per region.
• Reviews of one and two stars at your competitors. The complaints in them are a subject list. Every recurring complaint is an article or a question you answer better yourself.

In addition you measure the most important pages of three competitors with the same check script as your own pages. Then you compare figures instead of impressions. Which three those are you do not have to think up: the citation test from chapter 9 delivers the list.

The tooling: the free route first

Paid tools are good products. They pay off when time is more expensive than money or when you process hundreds of pages regularly. Not for this. Brand names we deliberately do not mention: such a list is out of date within a year.

A week in practice

Monday morning, automated
0  Check open actions from last week: carried out? measured? lapsed?
1  Fetch search data and behaviour data, keep a snapshot
1b Check own key pages technically, report deviations only
1c Enter changes of the last two weeks from version control
   as an event
2  Draw up the report: key figures, three signals, three subject chances
3  Check the advertising account: costs, search terms, warnings
4  Only the first Monday of the month: competitor block and self-check
4b Once a month, fixed date: the citation test from chapter 9
4c Once a quarter: lay the external profiles from chapter 10 next to
   the site
5  Record at most three actions. Reason obligatory

After that by hand
   Carry changes through, have them approved, publish

Step 0 is the decisive part. The rest is data gathering that exists everywhere.

Doing it yourself or outsourcing

Autocomplete asks this question literally, so here the honest answer.

Do most of it yourself. Switching on Search Console, looking for half an hour a week, writing down three actions and checking them the week after: that is not craft but persistence. Anyone who outsources it because he thinks he cannot do it often buys something other than he thinks.

Outsourcing pays off in three cases. You have hundreds of pages and a check by hand is not sustainable. The technical work touches the build of the site and there is no developer in house. Or nobody at your place keeps up the weekly rhythm: then you buy discipline and that is a valid reason, as long as you know that you are buying that.

What you never outsource is the decision what gets measured and which threshold justifies an act. Writing the scripts is the cheap part. SEMANU runs this loop on its own site and the baseline measurement you see in chapters 6 and 9 comes from it.

If you do buy it in, then ask three questions.

• Show me the action list of last month with the statuses added, not the report. Anyone who cannot show checked actions sells reports.
• In whose account are Search Console and the analytics package? That ought to be your own company account and the supplier ought to be a user there. Otherwise your history leaves together with the supplier.
• What happens in a month without result? The right answer is that it goes into the report that way.

Rates we do not mention here. They run too far apart to put an honest figure on them.

What you can do yourself and what you cannot

To do yourself, without a subscription:

• Create Search Console and confirm the ownership. Twenty minutes when you can get at your DNS.
• Half an hour in the performance report every week. Result: three actions.
• Keep the action list in one text file. No tooling is needed for it and a spreadsheet is enough.
• The monthly competitor block. Twenty minutes.
• The citation test from chapter 9. Count on an hour a month.

What you need help with or what you had better leave:

• The technical check script. About two hundred lines of code and one-off work for someone who programmes.
• Setting up your analytics package with goals that mean something. Wrongly set up goals are worse than no goals, because they look reliable.
• Drawing conclusions from weekly figures at low traffic. Leave that.
• Automatic publishing on the basis of a figure. The approval step is not a weakness of the system but its quality control.

Would you rather not do this yourself: SEMANU carries out the full SEO and GEO audit for 1,450 euro, report and working session included. What it covers is on the guide page.