Is my SEO agency working?
Ask for the changed URLs, verify what is live, and compare the affected pages in your own Search Console. Then check qualified enquiries or sales. These answer three different questions: was the work delivered, did search performance change, and did it help the business?
A green audit score cannot answer all three. Use this checklist at your next agency review, even if you never run an AuditLamp scan.
Check my site's technical evidence ↗
A free scan checks visible technical conditions. It does not judge your agency's work.
Download the free evidence worksheet · CSV spreadsheet · no signup
Check one claimed change from beginning to end.
Choose a page tied to a service, product or enquiry that matters to your business. Copy the exact task from the invoice into the worksheet. Ask the agency to fill the evidence columns before your meeting.
- Name the agreed task. Record the exact page URL, the problem, what was promised and the expected benefit. A title change, a redirect and an earned mention need different evidence.
- Ask for a dated before-and-after. Request the previous state, the deployed change and its release date. A draft or source-code commit alone does not prove a change reached your site.
- Verify the relevant condition. Check the current response, markup or rendered content. For a canonical claim, inspect the declared canonical, then use Google's URL Inspection for its selected canonical. Those are separate observations.
- Compare search outcomes on the affected page. In Search Console, use comparable complete periods, the same search type and the same device/country filters. Record clicks and impressions, not just the account-wide average position.
- Check the business outcome. Compare qualified enquiries, booked calls or sales with the agreed goal. If the forms, call tracking or attribution are missing, record that gap rather than inventing a conversion rate.
- Choose the next action. Mark the work verified, verification missing, outcome inconclusive or outcome improved. Set an owner and a review date. A failed outcome is not automatically proof the work was never delivered.
The worksheet has one row for each check and blank evidence fields. It contains no customer data or macros. Keep a separate copy for each review period.
Match the evidence to what you paid for.
| Agency claim | Evidence to request | What it establishes |
|---|---|---|
| We fixed canonical errors | The affected URL, its old/new canonical, response status and a current URL Inspection | The declaration changed; Google's selected canonical is a separate check |
| We improved the content | The old/new page, intended query, added information and publication date | A useful change was delivered; subsequent search results still need measurement |
| We fixed broken links | The source page, linked URL and current HTTP response or replacement link | The specified link now works; it does not establish a sitewide ranking gain |
| We improved performance | The same URL/device and comparable field or lab measurements, clearly labeled | A measured performance change; a lab run is not field Core Web Vitals |
| We earned relevant coverage | The live referring page and context, plus referral or search outcomes where available | A placement exists; a domain count alone does not prove customers arrived |
| We improved AI visibility | The named engine, query, date, observed answer and citation, with repeat checks | That observation occurred; crawler access or a paragraph format alone cannot prove citations |
Use the canonical checker, redirect checker or schema checker for a specific technical claim. Optional schema can be absent on a valid page. Its absence does not establish that an agency failed its contract.
Find which pages gained or lost useful searches.
Open Performance → Search results. Choose a comparison with equal, complete date ranges. Twenty-eight days against the previous 28 is a practical starting point; seasonal businesses should also examine the same season in the prior year. Keep the filters unchanged between periods.
Open the Pages table and select the page the agency changed. Review its clicks and impressions, then the available queries. Separate known brand searches from relevant nonbrand terms. Some queries are omitted for privacy, so the visible query rows may not add up to the totals. Leave that remainder unclassified.
Do not interpret a better sitewide average position as proof of recovery. The mix of queries can change. If exposure falls, poorly ranked queries can disappear and improve the average of what remains. Google's traffic-drop guide explains how to investigate affected pages, indexing, demand and other causes.
If your chart moves after a release, record the timing without declaring causation. Other changes, competitors and search demand can affect the same period. Save the export and your filters so the next review uses the same basis.
A verified repair can still have an unproven result.
Illustrative example, not a customer result: an invoice says a service page pointed its canonical to the homepage. The old HTML capture confirms that declaration. The new capture names the service page itself. The release date and Google's later inspection are recorded separately.
The worksheet can now say the canonical declaration was repaired. If no comparable search window or enquiries are available yet, the outcome remains unmeasured. Do not turn that technical pass into a claim that traffic increased.
For ongoing checks, save a baseline before work begins. AuditLamp's dated scan comparison compares captured fields from two saved reports. It cannot reconstruct an uncaptured past state or prove who made a change. Identical screenshots also do not rule out changes in metadata, server responses, off-site coverage or other work.
A 90-second check on your own homepage
You paid for work. The question is not whether someone was busy. The question is whether your site changed in ways you can verify. Most SEO changes leave evidence on public pages. You can inspect that evidence yourself.
- Seconds 0–20: can crawlers reach the site? Paste your URL into the free scan. Check whether search crawlers and AI answer engines are allowed to reach the pages you care about. A blocked crawler is evidence about access policy, not proof of lost citations.
- Seconds 20–45: are key pages labeled? Review titles, descriptions, headings and structured data on the pages your agency claims to have improved. Compare them to what the invoice described.
- Seconds 45–70: is primary content present without JavaScript? Check whether the main text exists in the captured HTML. A short preview alone does not establish a ranking or citation outcome.
- Seconds 70–90: read the worst finding first. Use the ranked findings as an itemized list of claims to verify with your agency. An open finding is a question, not automatic proof of bad faith.
A clean scan means the measured on-page conditions look sound. It does not prove rankings, traffic or AI citations. An ugly scan is not proof of dishonesty; it is a specific list of issues to discuss, including scope and timing.
Accountability scorecard (ten questions)
Use two sources you control: your Google Search Console property and a free AuditLamp scan. Download the evidence worksheet CSV if you want a printable record. No named agencies. Facts only.
| # | Question | Good answer looks like |
|---|---|---|
| 1 | Can Google and the named AI crawlers reach the site right now? | Agency points at a live scan row, not only a slide. |
| 2 | How many clicks 90 days ago vs today (Search Console)? | A number from Search Console, not only a proprietary visibility index. |
| 3 | Which pages are closest to useful search demand? | Named URLs, not “we are optimizing.” |
| 4 | What changed on the site last month (URL + before/after)? | Specific pages. Diffs when they have them. |
| 5 | Does the homepage state what you sell and where? | Matches what the scan flags under titles and answer blocks. |
| 6 | Are location or service pages unique, or copy-paste? | Scan may flag thin or near-duplicate patterns when present. |
| 7 | Is NAP (name, address, phone) consistent on the site? | One phone and one address string for the same location. |
| 8 | Core Web Vitals / mobile: what does real data show, if available? | Honest status, not only lab vanity. |
| 9 | If AI Overviews affected clicks, did they discuss it? | A conversation with evidence, not silence. |
| 10 | If you stopped paying tomorrow, what permanent assets remain? | Structure, schema and real pages — not only monthly posts. |
Open the ranked fix list. For each of the top five findings, write what broke, whether your agency already logged it this month, and what “done” means this week. A green agency report beside open scan fails is information and a reason to ask better questions — not automatic proof of fraud.
What should I ask my SEO agency?
What if they give me a green report but no more enquiries?
Ask which outcome each green item measures. Verified technical work and flat enquiries can both be real. Review the intended audience, search exposure, clicks and conversion tracking separately before deciding which part needs to change.
Does falling traffic mean my agency is doing a bad job?
It is a reason to investigate. It is not a diagnosis by itself. Ask for the affected pages and queries, comparable periods, an explanation supported by evidence and a bounded next experiment. Google's guidance on working with an SEO helps you evaluate expectations and recommendations.
Can AuditLamp prove that my agency did nothing?
No. We inspect available page conditions and compare saved observations. Work can be outside the scan's coverage, outside your agreed scope, not yet deployed or unrelated to page markup. Bring a specific finding to the agency and ask for evidence.
Will adding an answer paragraph get me into AI results?
A clear answer can help readers. It does not guarantee an AI citation or featured placement. Google's AI search guidance describes publisher requirements and the continuing role of foundational SEO.
Is the worksheet free? What does the scan cost?
The worksheet is free without an account. The website scan offers category measurements and selected findings without signup, with up to 150 pages and two fresh scans per domain per month. The $10 one-time Full Report opens detailed findings and export files. See current plan details.
Check a technical claim on your own site.
Enter your URL, review the scan's date and coverage, and take a specific finding into the agency meeting. The scan checks visible technical conditions, not your agency's performance. Keep the worksheet for changes and business outcomes that the scan cannot measure.
Free category measurements and selected findings. Full findings and export files: $10 once.
Sources and scope
Reviewed 24 September 2026. This checklist is a workflow for evaluating evidence, not a certification of an agency or a ranking prediction.