Using a Weekly Site Health Audit as an AI Crawlability Smoke Test

Technical problems rarely announce themselves. A template change, a plugin update, or a redirect rule lands on a Tuesday, and nobody notices until a report three weeks later shows a drop. By then the cause is buried.

A weekly audit narrows that gap. We use it as a smoke test: a quick, regular check that tells us whether anything has broken since last time.

What a weekly audit is for

It is not a deep technical review. A deep review happens at the start of an engagement and when something major changes. The weekly audit answers a narrower question: what is different since last week, and does any of it matter?

Promptwatch runs weekly Site Health audits with 21 checks, a health score, a summary of what changed since the previous run, and a weekly email. It also provides Site Content and Site Linking reports and crawlability checks. We use all of these, but the change summary is the part we read first.

Reading the change summary

We look at the delta, not the score. A health score that holds steady while two important pages quietly broke is not reassuring. Our reading order:

1. New issues. What appeared since last time? Each new issue gets a quick triage.
2. Resolved issues. Did our fixes land? Confirm they stayed fixed.
3. Affected pages. Are the problems on pages that matter to the client's prompts, or on low-value URLs?
4. Trends. Is the same kind of issue recurring? That suggests a process problem rather than a one-off.

We deliberately ignore a bulk of low-severity items. If every warning triggers a ticket, engineers stop reading our tickets.

The issues we escalate fastest

Broken or redirecting key pages. Pages that are cited or targeted for citations must return a clean response.

Indexing directives. An unexpected noindex or canonical change is high priority.

Blocked resources. Robots.txt changes or rules that affect bots.

Broken internal links to important pages. They cut off the paths bots follow.

Server errors and slow responses. They can lead bots to give up.

Each of these can silently cut a page out of AI answers, which is why a weekly rhythm helps.

Linking audits to crawler logs

An audit sees the site from a fixed vantage point. Crawler logs show what the bots actually experienced. We use both.

When the audit flags a problem, we check Agent Analytics to see whether real bots hit it. If bots hit the broken page and got errors, the issue is live and costing us. If bots rarely touch it, it is lower priority.

When the logs show errors, we check the audit to find the cause. A page that returns an error to bots may show a related issue in the site health checks, such as a redirect loop or a missing canonical.

Crawler logs need Professional or higher, or an agency plan, so this pairing is only available on those plans. For a client on Essential, we lean more heavily on the audit and flag the missing visibility.

The page-level view

Many issues are only meaningful in context. We cross-reference problem URLs with the Page Tracker and citation data. A broken URL that was once heavily cited is urgent. A broken URL nobody ever cited is housekeeping.

360 Insights joins crawler logs, citation clicks, and AI-referred visits per page, which helps us rank by consequence. Topic Insights highlights where pages underdeliver for their topic.

Building it into the rhythm

Our weekly sequence:

• Monday: read the audit email and the change summary.
• Monday: triage new issues into fix now, fix soon, or ignore.
• Midweek: send tickets to the client's engineers with the evidence.
• Friday: confirm any fixes and note results.

Items that need a client decision go on the Action Board with the owner and date, so they appear in the monthly review.

What we tell the client

We explain that the audit is an early warning system. We do not want them to worry about every flagged item. We send a short note on what mattered this week and what we are doing about it. When nothing needs attention, we say that in a line.

Avoiding false alarms

Some audit warnings reflect deliberate choices. A noindex on a thank-you page is fine. A redirect from an old URL is expected. We record such exceptions so they do not resurface every week. Clear notes save time.

What audits cannot tell you

An audit cannot tell you whether content is good, whether engines will cite it, or whether a competitor is ahead. It tells you whether the door is open. We need the other tools for the rest.

After migrations and launches

We run an extra audit immediately after any migration, redesign, or CDN change, then watch the logs for a few days. Migration problems tend to show up there first.

Our recommendation

Run a weekly audit, read the changes first, and pair the findings with crawler logs so you fix what bots actually hit. We do this with the Site Health audit and Agent Analytics in Promptwatch. Take a look at promptwatch.com.