Approving work you do not fully understand

Nobody expects you to evaluate a schema implementation. You do not evaluate your electrician's wiring either. You check that they explain the job, that it is permitted, and that the lights work afterward.

The failure is not ignorance. It is silent approval, where you sign off because asking feels like admitting something. That habit is how people end up with a site they cannot explain and a bill they cannot question.

Do I actually need to understand it?

You need to understand the decision, not the mechanism.

The mechanism is their job and you are paying for it. The decision is yours, because you are the one carrying the consequence. Those are separable, and confusing them is why owners either rubber-stamp everything or micromanage things they should not touch.

There is a useful test. If it went wrong, would you be able to describe to a new vendor what was done and why? If not, you have approved something you cannot own.

The point of asking is not to check the work. It is to make sure someone can explain it. Work that cannot be explained plainly is usually either poorly understood by the person proposing it or not worth doing.

What are the four questions that work on anything?

Ask them in this order and they cover nearly everything.

What problem does this solve? A real answer names a symptom you recognize. A weak answer names a best practice.

How will we know if it worked, and by when? The answer should reference a number you already track. If nothing will change measurably, ask why it is being done now instead of later.

What breaks if this is wrong? Some changes fail silently and some take the site down. You are entitled to know which kind you are approving.

Can it be undone, and how long would that take? This is the question that matters most and gets asked least.

Add a fifth when you can. "Where can I see this myself once it is done?" A page you can load, a listing you can check, a report line you can read. Learning to read one artifact is worth a lot, and a rank grid is the best candidate, since a grid shows you something a single ranking number cannot and it takes ten minutes to learn.

What does question two look like on a real proposal?

Here is the arithmetic. Use your own numbers.

Say the proposal is a speed project on your mobile site, and the pitch is that faster pages convert better. Question two is how you will know it worked, and by when.

Start with what you have. Take a month of mobile sessions and your inquiry count. Nine hundred mobile sessions at a 3% conversion rate is 27 inquiries.

Find the published effect. Google, Fifty-Five and Deloitte studied 37 brands across roughly 30 million sessions and reported that a 0.1 second improvement in mobile load time was associated with retail conversions rising 8.4% and average order value rising 9.2%. Travel conversions rose 10.1%.

Apply it. An 8.4% relative lift on 27 inquiries is a little over two additional inquiries a month.

Now judge the proposal. Two extra inquiries a month may be an excellent purchase or an obvious no, depending on what an inquiry is worth to you and what the work costs. Either way you now have a number to argue about instead of a feeling.

Then check whether the premise even applies to you. Google's Core Web Vitals thresholds are public: Largest Contentful Paint at or under 2.5 seconds, Interaction to Next Paint at or under 200 milliseconds, Cumulative Layout Shift at or under 0.1, all assessed at the 75th percentile of real visits. Run your own URL through PageSpeed Insights before you approve anything. If your Largest Contentful Paint is already at 1.9 seconds, there is no tenth of a second to buy back and the proposal is solving somebody else's problem.

And state the limits of the evidence out loud, because a good vendor will do it first. That study is October and November 2019 data. The brands were retail, travel, luxury and lead generation, not local service businesses. It reports association across a panel of large brands, not a controlled test on your site. It supports the direction. It does not forecast your result. A recommendation that arrives with its own caveats attached is a better sign than one that arrives with a bigger number.

Which decisions deserve a second opinion?

The irreversible ones, and there are fewer than you think.

Anything touching the domain. Moving registrars, changing the primary domain, redirecting the whole site. These are recoverable in theory and expensive in practice.

Anything changing your business identity on your profile. Name, address, primary category on an established listing, or anything that could trigger a reverification. The downside is being unverified for weeks during a dispute.

Anything that deletes or restructures existing pages at scale. A site migration or a URL change is a legitimate thing to do and it needs a rollback plan and a mapping document, in writing, before it starts.

Anything involving a third party writing on your behalf, including outreach and review solicitation. This is the category where the exposure genuinely lands on you rather than on the vendor, and it is worth being specific about why. Google's Maps content policy prohibits merchants from asking staff to collect a set number of reviews and from asking reviewers to include specific content, including content naming a staff member. Under the FTC's rule on consumer reviews, 16 CFR Part 465, an owner whose staff review the business without disclosing the relationship can be liable for the omission itself, at section 465.5. The agency writes the campaign. The business profile that gets a consumer alert banner is yours.

Everything else, a new page, a rewritten description, a post, an added service, is reversible in an afternoon. Approve those fast. Slow approvals on low-risk items are the main reason engagements stall, and it is usually the client, not the agency.

What if the thing I am being asked to approve is already obsolete?

This happens more than people expect, and it is the single best use of question one.

Two live examples you can check yourself in a browser.

FAQ schema sold as an AI or rich-result play. Google announced in August 2023 that FAQ rich results would only be shown for well-known, authoritative government and health websites, and that for all other sites the result type would no longer be shown regularly. A September 2023 update to the same post retired How-to rich results on desktop entirely. Google's current AI guidance also states that structured data is not required for generative AI features and that there is no special markup you need to add. FAQ content on a page is still useful. The markup has not produced the rich result for an ordinary local business since 2023.

Geo-tagged photos sold as a local ranking tactic. In Whitespark's 2026 Local Search Ranking Factors survey, 47 local search practitioners scored 187 factors. Presence of geo-tagged photos scored lowest of all of them. That is expert opinion rather than a controlled test, and it should be graded that way, but it is the strongest available signal on a tactic that is still being sold.

There is a third worth knowing because it carries actual risk. A plugin that scrapes your Google reviews and marks them up as an aggregate rating on your own homepage violates Google's structured data guidelines, which prohibit marking up reviews about your own business on your own site. It is one of the most common schema errors on local business websites, and it is a manual action risk with no upside.

None of these are proposed dishonestly. They are inherited from templates and decks that were correct four years ago. Question one catches them without anyone having to be accused of anything.

How do I ask without slowing everything down?

Use one sentence and stop apologizing for it.

"Explain this the way you would to my bookkeeper." That is not an admission of anything. It is a request for the version without jargon, and any competent practitioner has that version ready.

If the answer includes a claim about results, ask to see the measurement in your own account rather than in their dashboard. This matters most for newer areas where the tooling is immature. Traffic from AI assistants is the current example, because what shows in your own analytics for assistant referrals is limited and worth checking directly before anyone reports on it.

Know what your own account can and cannot tell you, because it saves an argument. Google Search Console added generative AI performance reports in 2026, and they give impressions, pages, countries, devices and dates. They do not give clicks, click-through rate, position, or the prompt a person typed, and AI Mode is not separated from AI Overviews. Clicks from those surfaces arrive in your analytics as ordinary Google organic traffic with no marker on them. So a report claiming a specific click count from AI Overviews is reporting something the platform does not publish.

If the answer includes a promise, slow down. Google states in its own documentation that no third-party tool has access to its internal ranking or AI systems, which is a useful sentence to have ready when a dashboard is described as showing what Google thinks of you. Certainty about outcomes is the one thing nobody in this field can honestly offer, which is why a guarantee should raise your suspicion rather than lower it.

And if you get an explanation that leaves you more confused, ask once more, then ask for it in writing. A second confusing answer is a signal about the explainer, not about you.

What to do this week

Take the last three things you approved. For each one, write what problem it solved and how you would know if it worked.

Whichever you cannot answer becomes a question. Send all of them in one email, framed as wanting to understand your own program better, which is exactly true. The replies tell you a lot about how the account is being run.

Then set a rule for yourself and share it. Reversible changes get approved within two business days. Irreversible ones get a written rollback plan and a week. Putting that in writing speeds your vendor up and protects you at the same time.

When you are evaluating a new agency, ask a reference specifically how the agency explained things. It is one of the more useful questions available, and reference calls yield real information when you ask about the working relationship rather than about results.

Questions about something you were asked to approve?

Email me at eric@seod.com with the proposed change pasted in, technical wording and all, and I will send it back translated into what it does, what breaks if it is wrong, and whether it can be undone. Plain language, no opinion on the agency proposing it, and no message afterward.

There is more on hiring and working with agencies if that is where you are.