First, check that both reports are for the same website and cover a similar number of pages. If the first audit checked ten URLs and the second checked twelve, the difference in the number of findings does not yet explain what changed on the pages included in both audits.
To verify a fix, match a specific URL with a specific issue. Review new pages separately. This will show whether the change actually worked and help you avoid missing parts of the site that were not checked before.
Check the conditions of both audits
For each audit, record the date, target, check type, pages actually processed, and known limitations. Distinguish the selected package from the result: a package for up to 100 pages does not confirm that all URLs were successfully retrieved. URLs that were not discovered and pages with processing errors need their own statuses.
Confirm the check method: the original response or the available state after rendering, the device mode if it is relevant to the measurement, and the settings applied. Results from different methods cannot be mechanically subtracted. A change in how a page is read can change the set of findings without any change to the website itself.
If the tool’s scope or evaluation method changed between runs, record this as a limitation of the comparison. You can compare an individual HTTP response, but the overall scores may now address different sets of questions. Do not call these scores fully comparable without confirming the conditions.
Create a list of URLs
Put the URLs from the two runs in two columns. Keep the original values alongside the working values: you may need to establish which address was actually requested during an investigation. Correct obvious copying errors, but do not remove all parameters or language prefixes just to make the strings match.
In the sample catalog, /chairs/?color=blue and /chairs/?color=red may show different sets of products. Combining them without checking would make the comparison misleading. Similarly, a mobile subdomain, regional section, or product variant URL cannot be treated as the same page based only on a similar name.
For documents that have genuinely moved, record an explicit mapping from “old address → new address” and the reason for it. Checking the redirect helps confirm the route, while reading the destination page helps establish its purpose. A similar path alone does not establish continuity without checking the content.
Find the pages in common
Suppose the first sample crawl processed ten URLs and the second processed twelve. Eight addresses appear in both. Two from the first crawl are not included in the second, and four appeared for the first time. Comparable observations are primarily available for the eight pages in common.
This does not mean the other addresses are unimportant. New pages need an initial check, and missing pages need to be investigated. Their states simply cannot be used as evidence that a previous issue has been fixed without additional data.
| Sample group | What you can conclude | Next action |
|---|---|---|
| Eight URLs in common | The same available checks can be compared | Compare findings by page and condition |
| Two URLs only in the first run | Their state on the next run is unknown | Check discovery, limits, and the current response |
| Four URLs only in the second run | New observations are available | Assess them separately from previous fixes |
Compare the address and issue together
The key for a working row is the URL and the condition being checked. For example: “on product page A, the image link returns 404”. If the second run retrieves an accessible file at the corrected address on product page A, there is a basis for confirming the result for that page.
However, the disappearance of the general “broken images” section may mean something else: the page was not retrieved, images were not extracted, or the check was not run. Review the details and limitations. An empty list after a failure is not a positive result.
The condition must also match. An image response and the presence of a valid alt attribute are different checks; fixing one does not confirm the other. For defining and verifying the results of topic-related tasks, see the technical audit guide, but make the decision based on the specific observations from both runs.
Mark the part of the site that was not checked
Use separate statuses in the table: confirmed issue, condition confirmed as met, retrieval error, and not checked. Do not reduce everything to two colors. Otherwise, an unmeasured page will end up in the fixed column just because the new document does not contain the previous row.
For the two URLs excluded in the sample, first find out whether they are currently accessible. One may have been removed by agreement, while the other may still be active but no longer discoverable from the new menu. These are different events: the first requires checking what happened to the address; the second indicates a gap in the repeated sample.
If the task was to fix an active product page, its disappearance from the crawl does not close the task. Run a suitable follow-up check of that URL and save the result. Do not replace evidence with “the tool no longer flags it.”
Record the result of the fix
In the summary, show separately the confirmed fixes on pages in common, new issues on pages in common, new URLs, and unknown repeat-check status. This is more useful than a single difference in total findings. It explains which part of the change is related to the crawl’s scope.
A sample statement could read: “The same conditions were compared on eight pages in common. There is no repeat data for two previous URLs; four new pages were reviewed separately.” Then list the actual evidence without inventing a number of successful fixes.
If you show an overall score, explain its limited role. An increase in the score does not prove an increase in traffic, usability across all user journeys, or the quality of unchecked sections. Separating the conclusions helps choose audit priorities for the next cycle.
Check results are available in the website monitoring dashboard.