Automated SEO reports: what to automate and what not

An SEO report is two kinds of work stacked together. One is assembling numbers: pulling the same data from the same places, in the same shape, every month. Machines do that better than people, because they do not tire, skip a step or round a figure differently in March. The other is judging what the numbers mean for one business: why something moved, whether it matters, and what to do. That needs knowledge no data source holds.

Google’s own advice is to check a tool’s recommendations before acting on them, as the guide to the SEO audit report sets out. A tool can count; it cannot know why a buyer chose a competitor this month, or which figures the business acts on.

An automated report is only as good as the data it pulls, and each source has limits that a dashboard does not show. Learn them once and write them into the report’s definitions.

Google sets out the Search Console limits in detail. The browser exports at most 1,000 rows; the API and the Data Studio connector reach up to 50,000 rows a day per site per search type, with a default of 1,000 rows per request that can be raised to 25,000^[1]. Queries issued by fewer than a few dozen people over two to three months are left out of the tables but counted in the chart totals, unless the report filters by query^[1]. The bulk export to BigQuery delivers all the performance data Search Console has for the property, except those anonymised queries^[2].

Timing matters as much as size. The API returns only finalised data unless it is asked for fresh data as well^[3], so a pull scheduled for the first of the month should ask for finalised data, or wait. In Analytics, rows can be withheld when too few users stand behind them, especially for search queries and demographics, and widening the date range can bring them back^[4]. Explorations and funnel reports keep event data for 2 or 14 months in a standard property, while the standard aggregated reports are unaffected by that setting^[5]. A report that needs three years of history must store it.

Data Studio, which Google called Looker Studio until recently, is the free way most teams turn Search Console and Analytics into a report that refreshes itself. Three of its behaviours decide whether an automated report can be trusted.

• Its Search Console connector reads one of two tables, site impressions or URL impressions, and one data source can use only one of them; to show both side by side, add two data sources to the report^[6]. The two tables count differently, so a chart and a table built on different ones will not agree.
• Queries that Google hides for privacy may appear as a single row with a blank value^[6]. Filter it out, or label it, so that nobody mistakes it for a query.
• Data from Google’s marketing and measurement products, including Search Console and Analytics, refreshes every 12 hours by default, and the Search Console rate cannot be changed^[7]. A report opened the morning after a fix may not show it yet.

Scheduled delivery sends the report by email as a PDF, with a preview of its first page and a link to the full report^[8]. Without the paid tier, each report can have one schedule, sent at most once a day; with it, up to 200 schedules and hourly sending^[8]. The schedule follows the time zone of the sender’s computer, which cannot be changed inside Data Studio, and a schedule that is deleted cannot be restored^[8].

The most useful automation is the kind nobody sees: tests that run before a report goes out and stop it when the data is wrong. A wrong report sent on time does more harm than a right one sent a day late.

1. 1Check that every source returned data for every day of the month, and that no day is preliminary.
2. 2Compare the month’s totals with the tool’s own interface once, by hand, when the report is built, and again whenever the pipeline changes.
3. 3Flag a metric that falls to zero or jumps several times over in a month. Most are tracking faults, not search events.
4. 4Watch the key events: if a form’s submissions drop to nothing on a day the site changed, the tag broke, not the demand.
5. 5Hold the definitions still: the same keyword set, the same pages in each group, the same attribution model. Log any change, with its date, in the report.

Some parts of a report are decisions, and a decision needs someone accountable for it. Keep these with a person who knows the business, and let the machine prepare the evidence for them.

• The cause of each change. A tool can show that clicks fell on a group of pages; it cannot know that the client paused a product line, that a competitor ran a sale or that the page was rewritten on the tenth.
• The label on each revenue figure. Whether a figure is known, influenced, modeled or incremental depends on the evidence behind it, as set out in the guide to SEO reporting in revenue, and a wrong label turns an estimate into a claim.
• The quality of the enquiries. Analytics counts form submissions; only the business knows which were customers, which were spam and which asked for something it does not sell.
• The next actions, and their order. Priorities depend on budget, people, dependencies and what the business needs this quarter.
• What is said about a decline. A fall explained badly loses more trust than the fall itself.

Language models can turn a table into fluent paragraphs in seconds, and the paragraphs will sound sure of themselves. That is the risk. A model reading a table sees that clicks fell; it does not know why, so it supplies the most common reason, and the most common reason is often wrong for this site and this month.

Use a draft, if at all, for the parts that describe the table: which rows rose, which fell, by how much. Then check every figure in the draft against the table it came from, delete every cause the draft supplied, and write the causes yourself from what you know was done and what happened in the market. Anything the draft calculated, such as a change or a share, is recalculated from the data, not taken from the prose.

1. 1Write the definitions first: the key events, the attribution model, the page groups, the keyword set and the brand terms. The automation encodes them; it does not choose them.
2. 2Pull Search Console through the API or the bulk export into storage you control, monthly at least, so that history outlives the tools’ own limits.
3. 3Pull Analytics with the same dates and the same time zone note, and join the two by landing page.
4. 4Build one dashboard with fixed views for the report, and keep exploration in a separate copy.
5. 5Run the checks before every send, and stop the send when one fails.
6. 6Schedule delivery for the first week of the month, after the person who owns the account has written the headline, the explanations and the plan.

What goes into the report itself, section by section, is set out in Monthly SEO report: what to include.