SEO Audit: What to Check, in What Order, and How to Read the Report

The audit PDF just landed. It runs dozens of pages, with an overall score on the cover and a long list of red alerts.

You flip through it, recognize half the words, and close the file with two questions. Is this serious? Where do I start?

The report leaves both unanswered. In it, the error that knocks your site off Google gets the same font as an image with no alt text.

We checked who answers when someone searches for this. On September 8, 2026, we measured the top ten organic results for “auditoria de seo” (SEO audit) on Google in Brazil, where we are based. Seven were agencies selling the service. The other three were tools selling a subscription. ¹ Not one of them taught you how to do it.

This piece does. You will leave knowing what an audit checks, in what order, and how to read the result. That helps you judge the report you received. It also lets you run your first audit yourself, with free tools.

What an SEO audit checks

Think of a home inspection before you buy a house. A good inspector starts with what kills the deal: the foundation, the wiring, the leak in the roof. The peeling paint comes at the end of the report, with the weight it deserves.

A report with three hundred items of the same size helps nobody decide. The crack in a beam sits next to a scuffed baseboard. You are left not knowing which of the two costs you the house.

The same rule holds for a website. Check everything in the right order, from what blocks the whole site to what is merely cosmetic. State the weight of every finding. This is called an SEO audit.

An SEO audit is not a list of defects. It is the order in which you decide.

Google itself writes what it expects from an audit, on the page for people about to hire an SEO. The audit should give you realistic estimates of improvement and an estimate of the work involved. If someone guarantees first place, Google’s advice is to find someone else. ²

The same page carries a line almost nobody repeats: “If you run a small local business, you can probably do much of the work yourself.” ²

The order of the exams matters more than the tool

Tools change names, screens, and prices. The order stays. In the book, I wrote that the button comes later, and that it changes names every six months. ³

The order below is mine. It adapts to SEO the six-exam protocol I published in the book to measure whether AI tools cite you. ³ Each exam only makes sense if the previous one passed. Skipping one is like reading a blood test run on a machine that was switched off.

The baseline exam: is the instrument measuring?

Think of the warning light on your car’s dashboard. You drove all year without seeing it come on, and you concluded everything was fine. The bulb had been burned out since the day it was installed. ³

Checking the instrument before reading the number is the first job of any diagnosis. Almost nobody does it. I call this the baseline exam. ³

The strangest case on my desk was a site with not a single row in Search Console for 140 days in a row. Over the same period, analytics recorded normal organic traffic. The first month without data was the biggest traffic month in the entire series. ³ With only one source on screen, the report would read as a catastrophe. With the other, a party.

It can go the other way, and it happened to me. My own book’s website stayed live for about 19 hours without recording a single session. A change in how the site was published left the measurement tag out of the code, and no alarm went off. That same day, another site of mine, with the same traffic sources, measured normally. ⁴ The dashboard showed zero, and the visitors kept coming.

The rule I take from those two cases fits on one line. Never diagnose from a single source. ³

Google says something similar when it compares its own two tools. Search Console clicks and Analytics sessions never match exactly. What matters is that the trends have the same shape. The same page reminds you that some pages on your site may be missing the Analytics tag. ⁵

Three checks close out the baseline exam:

1. Do Search Console and Analytics show the same trend over the last three months?
2. In Analytics, go to Admin, then Data display, then Events. Which events are marked as key events? Google describes them as the actions that are important to your business. ⁶ Page views, generic clicks, and scrolls fire on their own, without anyone deciding anything. If one of them is marked, the conversion rate in the report measures something else. ³
3. Does the Search Console Data anomalies page record any failure in your period? It lists the known data collection issues from recent months. ⁷

Exam 1: is something blocking the whole site?

Now the beam. Two Search Console reports answer this in seconds. Both show a green check mark when nothing is wrong.

The Manual actions report tells you whether a human reviewer at Google found pages on your site that break its spam policies. With a manual action, part of the site or all of it stops showing up in results. ⁸

The Security issues report tells you whether Google found hacked content, malware, or pages that deceive visitors. The affected page may get a warning label in search results, or a warning page in the browser. ⁹

At the same level come the three minimum technical requirements Google publishes. Its crawler, Googlebot, cannot be blocked. The page must respond with a 200 status code, the success code. And the page must have indexable content. Meeting all three makes the page eligible, and Google warns that indexing is still not guaranteed. ¹⁰

Exam 2: are the pages that sell in the index?

A page outside the index competes for nothing. This exam is covered in full in the piece on how to get your site on Google. It walks through the four doors a page passes through, plus the ten-minute exercise. Here is the short version.

The Page indexing report shows how many pages made it in and why the others were left out. Google’s help says all of your important pages should be indexed. It also tells you to check your homepage and key pages one by one, in the URL Inspection tool. ¹¹

During the inspection, also look at the Google-selected canonical field. It shows which version of the page Google chose to keep. ¹² If the chosen version is not yours, that is where the work starts.

Two tools round out this exam, and each one has a piece of its own. Screaming Frog crawls the site the way a bot does and shows what is inside it: response codes, titles, H1s. The Crawl stats report in Search Console shows what Google did on that same site. Reading it by response and by host is covered in the AI-powered audit. Reading it by purpose is in the piece on crawling and crawl budget.

Exam 3: does the page work for the people who land on it?

A store with a door that sticks loses the customer who was already on the sidewalk. The third exam looks at the experience of whoever opens the page.

Start with mobile. Google reads your site through its mobile version, crawled with the smartphone agent. That is the version it uses for indexing and ranking. Google calls this mobile-first indexing. ¹³

Next comes the speed that visitors actually feel. Google measures three things with data from real users. One is how long the largest element on the screen takes to appear. Another is how quickly the page responds to a tap. The last is how much the page jumps around while it loads. ¹⁴ That trio is called Core Web Vitals, and Search Console has a report just for it.

The help page for that report gives three reading tips. A group of pages gets the status that 75% of visits reached over the last 28 days. A site with little traffic may not have enough data, and then you get the no-data screen. And the order of repairs comes written out: first everything labeled Poor, then what affects the most URLs or your most important URLs. ¹⁴

Keep the limit in mind too. Google writes that good results in this report do not guarantee the top of search. It adds that chasing a perfect score just for SEO may not be the best use of your time. ¹⁵

Last come the labels. On a recipe page, a few lines of code can tell the machine the ingredients, the cooking time, and the calories. ¹⁶ This is called structured data.

The audit checks whether those labels are valid, in the rich result reports in Search Console. An invalid item has at least one critical issue, and that issue keeps it from appearing as a rich result. ¹⁷ Even with correct markup, Google does not guarantee the rich result will show up. ¹⁸ Mark up your pages for the reason the book defends. The label states clearly what the page and the company are. That way the machine does not mistake you for something else. ¹⁹

One detail separates an up-to-date report from a copied one. Since May 7, 2026, FAQ rich results no longer appear in Google Search. ⁷ An audit that still tells you to mark up FAQs to stand out on the results page is following an expired manual.

Exam 4: does the site show up, and for which questions?

Only now does the Performance report come in. Without the earlier exams, its chart shows the effect and hides the cause. It gives you clicks, impressions, click-through rate, and average position. ²⁰

Two readings pay off more than the chart. The first is to filter by page and look for high impressions with low clicks. Google describes that pattern. When impressions hold steady and clicks drop, the title and snippet may not be persuading anyone. ²¹ The same diagnosis, run in Bing’s dashboard, is in the piece on Bing Webmaster Tools.

The second reading needs a caveat. The queries list shows only part of your clicks. On the sites I measured, it covered between a quarter and two-fifths of actual clicks. The pages list covered all of them. ³ Pages measure volume. Queries measure intent. The two never go into the same calculation.

With the pages list in hand, open the ones that get the most visitors and read them as a visitor would. Google publishes self-assessment questions for this moment. One asks whether the content provides original information, reporting, research, or analysis. Another asks whether readers leave feeling they had a satisfying experience. ²²

Exam 5: did the visit turn into business?

The fifth exam only exists if the baseline exam passed. With a real key event set up, the question gets simple. Do the pages that get visitors make those visitors act?

Page by page, three questions settle it. How many people arrive here. What they do next. And what this page exists for. The piece on Page Analytics shows that reading, with the data laid over the page itself. It also lists what to check before installing the extension.

When you read a rate, two rules from the book apply. Use the rate of sessions that converted. Dividing the event count by sessions always inflates the rate, because a session with three events counts three times. And put what you are comparing against in the same sentence as the number. ³

Exam 6: what doesn’t the report know?

The last exam is a written list of what stayed unanswered. What the person searched for, when the query does not show up in the report. How much money each visit brought in, when the event has no value attached. What would have happened without the work. ³

Writing that list makes the report stronger. This exam, as the book puts it, marks the line between a diagnosis and a sales pitch. ³

How to read the result

A report in the right order still has to be read. Three rules settle almost everything.

Severity times reach. A defect is worth the size of the damage and the number of pages it hits. A block on the ten pages that sell goes ahead of an alert on three hundred archive pages. Google uses this logic in the Core Web Vitals report, when it tells you to prioritize your most important URLs. ¹⁴

A tool’s score is a model. Google writes that third-party tools do not have access to its internal ranking data, and that their predictions are their own. ²³ Take the Site Health score in Semrush’s Site Audit. It comes from the number and type of errors and warnings found in the crawl. It does not depend on how many pages were crawled. And a check you exclude drops out of the calculation. ²⁴ None of these criteria knows which page pays your bills. Use the score to compare the site with itself. Confirmation comes from Search Console, which Google recommends whether or not you use a third-party tool. ²³

A drop calls for three questions before any fix. The first is whether the instrument changed. Google’s guide to traffic drops tells you to check the Data anomalies page. Among the possible drop patterns, it even includes a reporting glitch. ²¹

There is one signature worth memorizing today. Impressions plunge, average position gets worse, and clicks do not move. In the book, I measured that step change on four unrelated sites, on April 20 and 21, 2026. I could not determine the cause. ³ This is called a series break.

Today the Data anomalies page records a case with that shape. From May 13, 2025, to April 27, 2026, a logging error kept Search Console from reporting impressions accurately. Only impressions, click-through rate, and average position were affected. Clicks were not. ⁷ The recorded end date is April 27, and the step I measured falls on the 21st. That is why I do not merge the two events. The rule still stands: before you diagnose a drop, check whether Google recorded a failure in that period.

The second question is whether revenue fell too. If both curves fall together, the customers went away. If only traffic falls, someone else is answering in your place. I call this the revenue test, and it costs one phone call to accounting. ³ Google’s guide gives the first case another name: seasonality and shifting interest. It suggests Google Trends to check. ²¹

The third is whether it is worth touching anything. Google recommends avoiding radical changes to a page that is already performing well, because small ranking fluctuations happen all the time. ²¹ In the book, the case that taught me this ended with the right answer being to wait. ³

How to evaluate the audit you received

With the order in your head, the PDF from the opening looks different. Read it again with these questions beside it:

1. Does the report start with the instrument, or does it open straight onto the list of alerts?
2. Does every number come with its source and date?
3. Are the findings in order of impact, or in the order the tool prints them?
4. Does the report say what is happening to the pages that sell?
5. Does it separate Google’s data from a tool’s alerts?
6. Do the recommendations cite official documentation? Google suggests asking exactly that. ²
7. Is there a section saying what could not be known?

Two of Google’s own rules apply when you hire. Give the auditor read access to Search Console, with no write access at this stage. And check the recommendations against official documentation before making big changes based on a tool’s audit. ²

Your first audit, in one afternoon

You need four free tools. Search Console, Google Analytics, Bing Webmaster Tools, and the free version of Screaming Frog, which crawls up to 500 URLs. ²⁵ You also need a sheet with four columns: finding, affected page, source, and date.

Follow the order, without skipping:

1. Baseline exam. Compare the trend of Search Console clicks with organic sessions in Analytics over the last three months. List the key events and the volume of each one. Open the Data anomalies page.
2. Exam 1. Open Manual actions and Security issues. Write down the green check mark, or the problem.
3. Exam 2. Write down the ten pages that sell. Inspect each one and note the ones missing from Google, with the reason given.
4. Exam 3. In the Core Web Vitals report, mobile view, note the groups labeled Poor. In the rich result reports, note the invalid items.
5. Exam 4. In the Performance report, choose the last 16 months and the pages table. Google’s guide recommends that window to tell a drop apart from seasonality. ²¹ Note the pages with high impressions and low clicks.
6. Exam 5. For the five pages with the most visits, answer the three questions.
7. Exam 6. Write down what stayed unanswered.

From now on, when an audit report arrives, you will look for the baseline exam on page one. If it is not there, your first question for the meeting is ready.

Open Search Console today, write the date at the top of the sheet, and start with the dashboard light.