Answer engines send their own crawlers — OAI-SearchBot, GPTBot, ClaudeBot, PerplexityBot, Google-Extended and others — and each can be blocked independently of Googlebot. Blocks show up in robots.txt, but they also hide in server configuration, security plugins, firewall rules, and CDN settings, none of which are visible from inside your website admin. A site can rank respectably in Google and be entirely unreadable to ChatGPT, and nothing in a standard SEO report would reveal it.

They name specific crawlers, and cite server access logs or live request testing as evidence. They can tell you which bots have actually hit your site and when.

A robots.txt screenshot presented as proof, or an unfamiliarity with the crawler names. Robots.txt is a request; a server-level block is a wall, and only one of them shows up in a file you can read in a browser.

This is the single fastest way to separate vendors who measure generative visibility from vendors who talk about it. It takes about four minutes to run live. Ask them to do it in the meeting, with the questions a real prospect would ask — not your firm name, but the problem your clients have.

They run it in front of you, show you which competitors get named, and flag anything the engines say about you that is inaccurate or outdated. Inaccuracy is a finding, not an afterthought.

Deflection to a proprietary "AI visibility score" with no underlying prompts shown. If you can't see the question that was asked, the number means nothing.

Some of this field is settled and testable. Some of it is well-supported pattern recognition. And a meaningful share of what circulates as generative-search best practice is extrapolation from a thin research base plus a lot of confident guessing. Those are three different things, and you are entitled to know which category you're funding.

Findings arrive labeled by confidence, and the vendor volunteers where the evidence is weak without being pushed. Ours use four tiers — verified, directional, pending, unconfirmed — but the labels matter less than the fact that some exist.

Every finding presented with equal certainty. A 200-item audit where nothing is marked uncertain is not thorough; it's undifferentiated.

This question tests incentive alignment more reliably than anything on their pricing page. A vendor billing hourly or on retainer has a structural reason to find more work. That doesn't make them dishonest — it makes them human — but it means the answer to this question is genuinely costly for them to give, which is exactly why it's informative.

A specific example from a real engagement. "We told a client to stop paying for directory submissions." "We recommended against a redesign they'd already budgeted." Specificity is the tell.

A generic answer about not doing black-hat tactics. Everyone says that. It costs them nothing to say.

When one firm both diagnoses the problem and bills for the solution, the diagnosis and the invoice have a relationship. Again: not necessarily corrupt. But it's a conflict you should price in, the same way you'd want to know if your home inspector owned the roofing company.

Either the roles are separated, or the vendor names the conflict openly and explains how they manage it — fixed scope, independent verification, a client-side developer who can push back.

Puzzlement that you asked. The conflict is real whether or not the vendor has thought about it, and a vendor who hasn't thought about it hasn't managed it.

There's a real distinction between visibility you own and visibility you rent. Structural fixes to your own site — schema, page architecture, answer-ready content, crawler access, entity consistency — persist after the engagement ends. Paid placements, syndicated content on the vendor's network, and links from properties they control disappear the month you stop.

A clear split: here's what stays, here's what stops, here's roughly what proportion each represents. A vendor who's thought about it will have the answer ready.

"Your rankings would drop." That's an answer about their leverage, not about what you bought.

Prioritization is the actual skill. Anyone can generate a list of two hundred issues from a crawling tool; the expertise is in knowing which three matter and which hundred and ninety are noise you should never pay to address. Asking for the first item forces a judgment they can't hedge.

One specific item, with a stated reason tied to your business rather than to a score — "you're invisible to two of the four major answer engines, so nothing else you fix can be seen by them." Sequencing logic, not severity color-coding.

"It depends, we'd need to do a full audit first." Fair as a caveat, but they've presumably looked at your site before this meeting. If they truly have no working hypothesis, they haven't prepared.

This one is boring and it ends more engagements badly than any of the others. Analytics properties, Search Console, your Google Business Profile, the domain registration, the hosting account, and the copy written during the engagement should all be in your name. Vendors sometimes create these under their own accounts for convenience, and convenience becomes leverage at exactly the moment you want to leave.

Everything in your name from day one, with the vendor added as a user. Content ownership stated in the contract. They offer this before you ask.

Accounts created under the agency, with a promise to transfer them later. Later is when the relationship is worst.