Site Audit Summary
A one-page report on what is wrong with a website and what to do about each thing, measured live and written to be exported as a PDF and sent to the owner.
Last updated 2026-10-03
Summary#
Site Audit Summary tests one page of a website on a phone and on a desktop, looks up how much organic traffic it gets and who links to it, and turns the result into a one-page report: what is wrong, what the measurement was, and what to do about each thing. It is written to be exported as a PDF and sent to the business that owns the site.
Purpose#
Telling a business its website is slow is easy and gets ignored. Showing it that 807 KB of image weight can be removed, with the test that measured it, is what gets a reply.
This exists to produce that document in about a minute instead of an afternoon, and to produce it honestly: every line names the check it came from, and anything we could not test is listed as untested rather than reported as a fault.
Overview#
Two live Google Lighthouse runs, one as a phone and one as a desktop, through the same PageSpeed Insights service behind pagespeed.web.dev. Nothing is cached, so the scores are current and the person receiving the report can reproduce them by typing the address into Google's own tool.
On top of that, two figures Lighthouse cannot see: estimated monthly organic traffic, and how many other sites link to the domain.
The findings are an edited list, not a Lighthouse export. A typical run has 30-odd failing audits, most of which are metrics such as "Time to Interactive" or diagnostics such as "Network dependency tree". Those are consequences of the real problems rather than separate jobs, so they are reported as measurements and kept out of the list of things to do.
Benefits#
- Every finding carries the number that failed it, so the report survives being checked.
- Both devices, because they disagree. A site scoring 60 on a desktop commonly scores 30 on a phone, and local customers are almost all on the phone.
- Every number the report quotes is text, so the whole thing exports and reads on paper without an image to chase.
- It states what it did not check, which is what makes the rest of it credible.
- Exports to PDF, Excel, CSV and JSON like every other tool.
Use Cases#
Opening a conversation with a prospect. The report is the pitch. It is specific, it is checkable, and it does not require the reader to trust you.
Qualifying a lead before you call. A site with a 30 performance score and no meta description has work worth quoting for. One scoring 95 does not.
Showing a client what has changed. Run it before a project and again after.
Checking a site you have just launched. The findings are the snag list.
Auditing an inherited site. One run tells you what the last developer left.
Requirements#
- A signed-in account.
- A paid plan. Any cost above zero is blocked on Free.
- A website address that Google can load from the public internet. A site behind a login, an IP allow-list or a holding page cannot be tested.
- Available credits and hourly headroom.
- No integration and no verified domain. It works on any site, not only your own.
Permissions#
Any workspace member on a paid plan can run it, under the Site Health grant. A teammate running it inside a workspace spends the workspace owner's credits.
Cost#
3 credits per site.
The two Lighthouse runs are free: Google provides them. The charge covers the traffic and backlink lookups and the platform work around them. At 3 credits it is at the premium threshold, so it counts against your monthly allowance rather than only the hourly fair-use limit.
A site Google cannot load is not charged. The run is refused with a message saying so, and the reserved credits are returned.
There is no cache on this tool, deliberately: pressing Run always runs a fresh test. See How credits work.
Navigation Path#
Dashboard, then Site Health in the left rail, then Site Audit Summary.
Inputs#
| Field | Required | Notes |
|---|---|---|
| Website address | Yes | The page to test, usually the home page. A bare domain such as acmeplumbing.com is fine; the https:// is added for you. |
One field, because the tool decides everything else. See Common inputs and validation.
Step-by-Step Guide#
- Open Site Health, then Site Audit Summary.
- Type or paste the website address.
- Press Audit this site. It takes about a minute, because two live Lighthouse runs are genuinely slow. The status line counts up so you can see it is still working.
- Read the findings, worst first.
- Press PDF in the export row to produce the report, and send it.
The explainer below the result walks through how to read each part of the report.
Reading the Results#
The report is the PDF. A run audits one website and produces a document you can send a client, so the screen shows a short summary and a Download the PDF report button rather than the report itself. The export row PDF button hands over the same file. Excel, CSV and JSON still export the same figures as data.
Inside the PDF:
Scores come first, on both devices, drawn as the same four gauges Google shows. Green is 90 and above, amber 50 to 89, red below 50, in Google's own colours for those bands, because the report should not contradict the thing it is quoting.
The screenshot Google took of the page on each device. It is the picture from the PageSpeed result, not a separate capture of the site.
How the site is doing in search carries four numbers: organic visits a month, keywords ranked for, sites linking here, and links in total. Any of them that could not be measured says not checked rather than showing a zero.
What is wrong follows as a numbered list, worst first. Each line is one short sentence and the number behind it, such as Images are not compressed: Est savings of 807 KiB.
The report does not say how to fix anything. That is deliberate, and it is an owner decision: the document states what is wrong and stops. The fix text is still carried in the JSON export for anyone who needs it.
Severity is about order rather than drama. Critical means nothing else on the list can help until it is fixed, which in practice means search engines are blocked from the page or the site is not served over HTTPS.
What this report did not check is the last section, and it is always present. See Reading a result.
Examples#
A real run against a roofing company's site returned 26 findings, including 807 KB of removable image weight, 2,637 KB of unused JavaScript, 3,918 KB re-downloaded by returning visitors because of cache settings, a total page weight of 19,701 KB, links that search engines cannot follow, and images with no alt text. Its scores were 30 on a phone and 60 on a desktop.
Screenshots#
Screenshots are not embedded in this article. The explainer below the result shows the layout and how to read each part.
Tips#
- Test the page you would send someone to, which is usually the home page but is sometimes a service or landing page.
- Run it on your own site first, so you know what the report looks like before you send one.
- The phone score is the one worth leading with for a local business.
- The gap between the two devices is often more persuasive than either number.
- Pair it with Local Leads: that finds the businesses, this says what is wrong with their sites.
Best Practices#
- Send the PDF rather than a summary of it. The measurements are the argument.
- Do not strip the "what this report did not check" section. It is what stops a reader treating the rest as a sales document.
- Re-run after work is done. A before and after is the easiest case to make.
- For a site you own and want to fix thoroughly, run Site Audit as well: this tests one page, that crawls many.
Common Mistakes#
- Treating it as a full site audit. It tests one page. Site Audit crawls the site.
- Reading the search figures as exact. They come from third-party index data, not the site's own analytics, so they are the right order of magnitude.
- Assuming a short list means a good site. It can also mean checks did not run, which is why the unchecked section exists.
- Testing a page behind a login. Google cannot load it, so there is nothing to measure.
Limitations#
- One page per run.
- No competitor comparison. The report used to carry a line naming the competing domains it found. That was removed on 2026-10-11 by owner decision, along with the provider lookup behind it, so no competitor data is fetched or printed at all now. Use Competitor Battle or Gap Finder.
- Lighthouse scores vary between runs. A score can move several points between two tests of the same unchanged page, because it is a live measurement on shared infrastructure. Treat a difference of a few points as noise.
- Google's testing service has a daily quota. In heavy use a run can be refused for that reason, which is reported as such.
- Estimated traffic is not analytics. It is modelled from ranking data.
- It cannot test a site it cannot reach, including sites behind a login, an IP allow-list or geographic blocking.
Troubleshooting#
"Google could not load that site." The address is unreachable, behind a login, or blocking automated requests. Nothing was charged.
The audit is taking a long time. A minute is normal. Two live Lighthouse runs against a slow site can take longer, and a very slow site is itself the finding.
The scores changed between two runs. Expected. Lighthouse is a live measurement; a few points of movement is noise.
A figure says "not checked". That provider did not answer. No conclusion is drawn from its absence, and the unchecked section says which one.
The list is shorter than I expected. Only checks that actually failed are listed. Nothing is added to pad it.
FAQs#
Where do the scores come from? Google Lighthouse, run live through PageSpeed Insights. Anyone can reproduce them at pagespeed.web.dev.
Why does it test twice? Because the results differ, often by a lot, and reporting only one would be choosing the flattering half.
Will it tell me a good site is bad? No. Only checks that failed appear, with the failing value beside them. A site with nothing wrong gets a report that says so.
Is this the same as Site Audit? No. Site Audit crawls many pages of a site and produces an internal work list. This tests one page, adds the search figures, and is written to be sent to somebody else.
Can I audit a list of sites at once? Not yet. One address per run today.
Does the report mention Metric Vault? The exported PDF carries your workspace branding, the same as every other export.
See also
Was this article helpful?
Thanks — feedback noted for the docs team.