Should You Run a Free Marketing Audit Before You Buy?

You are about to commit new budget and you are not certain the current spend is working. Yes, run the audit, but judge it by what it measures and what it hands back in writing, not by whether it costs nothing.

Free is not the interesting variable. Every agency on earth offers a free audit, and most of them are a sales call with a slide deck attached. The useful ones are diagnostics with a defined scope: they answer a specific question you already have, they use your data instead of generic benchmarks, and they end with a written finding your team can act on whether or not you ever hire the people who produced it.

That is the standard. Here is how to apply it.

The three questions an audit has to answer

Strip away the formatting and every marketing audit worth reading answers three things.

Where is demand actually coming from. Not what the ad platforms claim. Platform-reported ROAS and last-click reporting are diagnostics, not truth, and they double count by design. A real audit reconciles what the ad accounts say against what the analytics property says against what closed in the CRM, then tells you which sources produced revenue rather than clicks.

Where does it leak. Demand arrives and then something loses it. The leak is almost never where the dashboards point. It sits in checkout, in a form that routes nowhere, in a lead that waits most of a day for a reply, in a lifecycle email flow that was switched on two years ago and never touched since. Naming the leak with a number attached is the hardest part of an audit and the part most audits skip.

Which single fix moves revenue fastest. Not twenty-three recommendations ranked by effort. One. If the document hands you a list of everything that could be better, it has not done the prioritization work, it has transferred it to you.

If an audit answers those three questions, it earned its slot in your calendar. If it tells you your Core Web Vitals could be stronger and your meta descriptions are too long, you received a crawl report, not a diagnostic.

What access they need, and why read-only is enough

The usual reason a marketing leader refuses an audit is access. Nobody wants to hand admin rights to a vendor they have not hired.

You do not have to. Every platform that matters has a view-level role, and view-level is sufficient for diagnosis. The exact role names differ by platform and change over time, so confirm current naming in each platform's own help documentation before you grant anything, but the list of surfaces is stable:

• Ad accounts. Google Ads and Meta, at a read-only or analyst-equivalent permission level. Enough to see spend, structure, creative performance and conversion settings.
• Analytics. Your GA4 property at viewer level. The conversion configuration matters as much as the traffic numbers.
• CRM. Enough visibility to see lead source, stage progression, response times and close rates. This is where attribution either survives contact with reality or does not.
• Email and lifecycle platform. Flow structure, send volume, engagement trend and list health.
• Commerce platform. For Shopify brands, staff or collaborator access scoped to analytics and orders.
• Search and AI citation data. Search Console, plus whatever visibility measurement covers AI engines, which is the one most audits do not request because they do not look at it.

Read-only access protects you and costs the diagnostician nothing. Anyone who insists on admin before a proposal is telling you something about how they work.

Most audits still ignore AI search visibility

This is the largest blind spot in the standard audit scope, and it is no longer a forward-looking concern.

The assumption underneath most search audits is that ranking well on Google means being visible. That assumption has broken. A Search Atlas study by Manick Bhan, published 10 November 2025 across 18,377 semantically matched query pairs collected in September and October 2025, found 7 percent URL overlap with Google for OpenAI's GPT and 6 percent for Gemini, in a test setup where GPT and Gemini ran with WebSearch disabled. At the domain level the same study records 21 percent for GPT, 28 percent for Gemini and 43 percent for Perplexity. BrightEdge, in its AI Overviews one-year analysis covering February 2025 to February 2026, puts only about 17 percent of Google AI Overview citations in the organic top ten, so roughly five in six come from outside page one, ranging from 24 percent in healthcare to 11 percent in finance. And an unreviewed preprint, "Source Coverage and Citation Bias in LLM-based vs. Traditional Search Engines" (arXiv 2512.09483, December 2025), measured ChatGPT's mean per-response domain overlap at 0.118 with Google and 0.041 with Bing.

Read those together and the conclusion is uncomfortable but simple: visibility does not transfer cleanly. In that test setup, a Google ranking rarely earned a Gemini citation at the URL level, and the overlap between engines varies widely by engine. They are separate discovery surfaces with separate mechanics.

So an audit that measures rankings and stops is measuring a slice of discovery that is getting smaller while reporting it as the whole. If your buyers are building shortlists inside an AI engine, an audit that never checks whether you appear in those answers has left the most consequential question unasked.

When you commission an audit, ask one thing: what do AI engines say when someone asks for a recommendation in our category, and are we in the answer. If the response is a blank look, the scope is from 2021.

A finding names a number, a cause and a timeline. A pitch names a service.

Here is the test that separates a diagnostic from a sales document, applied to the artifact you actually receive.

A finding looks like this, as a hypothetical: paid social is credited with a large share of leads, but those leads close at a materially lower rate than organic search leads, the cause is a lead magnet that attracts researchers rather than buyers, and changing the offer would show up in close rate within a defined window.

A pitch looks like this: your paid social could be performing better. We offer performance marketing.

Number, cause, timeline. Three elements. One of them missing and you have an observation with a call to action stapled to it. All three present and you have something your team can execute on Monday with nobody else in the room.

Insist on the written artifact. A findings call with no document is designed to be unverifiable and unportable, which is the point. Written output can be forwarded, challenged, costed and handed to a different vendor. That is exactly why some audits never produce one.

The checklist to run on any audit you receive

Five questions. Run them on the document, not the presentation.

1. Does it use my data? Every number should trace to one of my accounts. Industry averages are context, not findings.
2. Can my team execute this without the people who wrote it? Check how many recommendations depend on the author staying on retainer, and treat heavy dependence as a sign you are reading a proposal.
3. What happens if I do nothing? A real diagnosis has a cost of inaction. If no one can quantify the downside of ignoring it, nothing was measured.
4. Is there one prioritized move, with a reason it is first? Sequencing is the expertise. A flat list is not.
5. Does the scope include AI search visibility, or only rankings? With URL-level overlap between Google results and AI citations in the single digits, and only about 17 percent of AI Overview citations ranking in Google's organic top ten, this is now a scope gap, not a nice to have.

An audit that survives those five is worth acting on. One that does not is a free hour you spent on someone else's pipeline.

Where we land on it

We run Discovery and Audit as weeks one and two of every engagement, across the full existing stack, because you cannot prioritize what you have not mapped. What comes out of it is a stack and integration map, a 90-day quick-win roadmap tied to specific revenue outcomes, a 12-month architecture, a tool stack rationalization plan, and a governance and measurement framework. It is documented, and the handoff plan assumes your team owns it afterward. That is the deliverable whether the relationship continues or not.

If you want that question answered about your own accounts, including what AI engines currently say about your category, book a strategy session.