How to check for yourself whether things are working

A client who cannot check for themselves what they are getting depends on the goodwill of the supplier. This chapter exists so that this is not necessary. It will not make you a specialist, but it gives you enough to ask the questions that an honest report answers easily, and a dressed-up one not at all.

Three numbers that say almost nothing

Marketing reports are full of numbers that look good and do not mean much. Three of them show up almost everywhere.

Average position

It is calculated as an average across all the searches in which you appeared, no matter how rarely or how irrelevant. An average position that climbs from 24 to 19 can mean that you have started to appear, in poor spots, for many more new searches. The number improves, the people do not come.

Number of impressions

It measures how many times a link to the site appeared in a list of results. It does not say whether anyone saw it, let alone whether anyone clicked on it. A site can rack up hundreds of thousands of impressions on searches that have nothing to do with what it sells.

“Total traffic”

It adds up everything: visitors from searches, from social networks, from paid ads, bots, your own visits. If a report shows you the growth in total traffic as proof of the optimization work, ask for the breakdown by source. The difference is often surprising.

Three numbers that say something

All three can be read for free in Search Console, the tool Google makes available to anyone who owns their site. If you do not have access to it, ask for it. It is yours.

What this looks like in practice, step by step

Search Console has many screens, but for the monthly check you need only one: the performance report. Here is exactly what to click and what to read.

1. Open Performance and set the date range to the last 3 months. A single month is too noisy to show a trend.
2. At the top, tick the Total clicks box and untick Average position. You are left with what matters on the screen.
3. Scroll down to the table at the bottom, the Queries tab, and sort by clicks, not by impressions. Those are the questions your site actually wins.
4. Switch to the Pages tab, again sorted by clicks. Count how many pages have at least one click a month. That number, tracked over time, is the most honest measure of your content.
5. Write down three things in your table: how many clicks in total, how many pages bring clicks, and the top five queries.

If the number of pages that bring people grows from month to month, the work is working, even if the average position has dropped. If that number stalls while impressions grow, you are producing visibility without relevance, and that is fixed through content, not through settings.

Checking in AI assistants, differently from Google

In a classic search engine you care about position. In an AI assistant there are no positions: either you are cited as a source in the answer, or you do not exist at all. The check is done differently.

Ask the question exactly as a client would phrase it, in a new conversation, without mentioning the company name. Then watch for three things: whether a link to your site appears, whether the information in the answer is correct, and, above all, which companies are cited instead of you.

The last point is the most useful. The sources cited by an AI assistant show you directly what kind of pages are considered trustworthy on that topic. Often they are not company sites but articles, comparisons or discussions elsewhere, and that changes where it is worth investing the effort.

A warning that spares you wrong conclusions: AI assistants give different answers to the same question, from one day to the next and from one user to another. A single check proves nothing. Three months of monthly checks, written down in the same table, start to show something.

The trap that matters most

Here we get to the part you will not find in other guides, because it is learned only by making mistakes.

This is not a hypothesis. We ran into it on one of our own sites, on the day we are writing this chapter.

If we had stopped at the first number, we would have reported 91 and we would have been right on paper. The site would have been left with all its internal links broken for search engines.

Hence the practical rule: before you read an audit's score, read how many pages it analyzed. A report that says “we analyzed 14 pages, 2 of them articles” is verifiable. One that only says “score 91 out of 100” is not information, it is an impression.

Why the same site gets different scores on the same day

Audit tools do not read the whole site, it would take too long. They pick a sample of pages. If the sample changes, the score changes, without anything having happened to the site.

On the same day, one of our sites showed 100 in the morning and 98 in the afternoon. Nothing had changed on it. The tool had picked different articles, and the ones picked the second time did not cite external sources. The second measurement was closer to the truth than the first.

The second reason for results that do not match is the cache. Fast sites keep copies of their pages so they do not have to rebuild them on every visit. After a change, a tool can receive the old copy and report that nothing has changed. It happened to us three times in a single day.

What to do every month, in twenty minutes

You do not need paid tools. You need a list of questions written once and the patience to repeat it.

1. Choose ten to twenty questions that a real client would type. Not your company name, but the problem you solve: “how much does X cost”, “how do I choose between X and Y”, “what do I do if Z happened to me”.
2. Run them month after month in Google, but also in an AI assistant such as ChatGPT or Perplexity. Each time, note whether you appear, in what position and, for the assistants, whether you are cited as a source.
3. Also note who appears instead of you. The competitor who shows up consistently tells you more about what needs to be done than any report.
4. Open Search Console and look at the queries that brought clicks, not at the average position.
5. Write everything in the same table, with the date. Three months of such a table are worth more than a fifty-page report.

Repetition matters more than refinement. A simple list, run identically every month, shows a trend. A complicated analysis, done only once, shows nothing.

What to ask from whoever does your marketing

Including from us. An honest report contains four things, and the absence of any one of them is a sign.

• What was concretely done this month, at the level of page and change, not “we optimized the content”.
• Numbers with the date of measurement. A value without a date cannot be verified by anyone, not even by whoever wrote it.
• What did not work and what was learned from it. A report in which everything went perfectly every month describes something other than reality.
• What comes next and why, in that order.

Sources

• Google Search Console documentation, the section on the performance report and on how average position is calculated.
• The numerical examples come from our own audits run on 18 August 2026 and are reproduced with the exact figures from that day.