The short answer. Hire the consultant who can show you how they will measure the work before they do it, who asks for access before quoting, and who declines to promise rankings, mentions or citations. Everything else on this checklist follows from those three.
This is written for a marketing lead or founder comparing proposals. It applies to SEO consultants, “AI SEO” consultants and agencies offering GEO or AEO; the questions are the same because, as we explain in SEO vs AEO vs GEO, the work overlaps.
Method
- “Walk me through the order you work in, and why.”
- “What will you tell my developers to change?”
Listen for: a sequence from access to measurement
Measurement
- “How will we know it worked?”
- “How do you measure mentions in ChatGPT or Perplexity?”
Listen for: named metrics, named sources, a baseline first
Claims
- “Can you guarantee a ranking, a mention or a citation?”
- “Which of your case studies can I check?”
Listen for: no, to any promise of a position
Access and process
- “What access do you need, and when?”
- “Who does the work?”
Listen for: report access before the quote
What should you decide before you talk to anyone?
Three things, written down:
- The symptom. Traffic fell; key pages are not indexed; assistants describe you wrongly; competitors appear in shortlists and you do not. A consultant who hears a precise symptom can give a precise plan.
- Who will do the implementation. Your developers, your writers, or the consultant’s team. This decides the shape of the engagement more than anything else.
- The decision you want to make at the end. “Fix this list in this order” is a diagnostic. “Run our search channel” is a retainer. Mixing them up is how reviews turn into open-ended projects.
What questions should you ask about method?
“Walk me through the order you work in, and why.” Good answer: a sequence that starts with access and indexing and ends with measurement, because later work depends on earlier work. Worrying answer: a list of tactics with no dependency between them, or a method that starts with content production before anyone has checked whether the crawler can read the site.
“Which parts apply to AI assistants specifically?” Good answer: a crawler-access check for the documented agents, an entity review, and a way to make passages quotable. Worrying answer: “AI SEO is completely different” (it is not; the engines retrieve from the same index) or “it’s all the same as SEO” (the measurement and the entity work are new).
“What will you tell my developers to change?” Good answer: ticket-ready items with the expected effect of each. Worrying answer: a PDF of screenshots.
What questions should you ask about measurement?
“How will we know it worked?” The single most useful question. Good answer: named metrics with a named source and a baseline taken before the work starts. For classic search, Search Console impressions, clicks and indexed pages. For AI surfaces, the two first-party reports that exist, Search Console’s Generative AI features report and Bing Webmaster Tools’ AI Performance report, plus a stated sampling method for the assistants that report nothing.
“How do you measure mentions in ChatGPT or Perplexity?” Good answer: a fixed prompt set, several runs per prompt, logged-out sessions, results reported as rates with the spread shown, and the prompt set published or at least shared. Worrying answer: a single screenshot of one answer, or a tool’s “AI visibility score” with no explanation of what it samples.
“What will the first report look like?” Ask to see a redacted example. If there is none, ask for the template.
What questions should you ask about claims?
“Can you guarantee a ranking, a mention or a citation?” The only acceptable answer is no. Google’s documentation on AI features describes no mechanism for buying or securing a place in an AI Overview, and no assistant vendor documents one either. A consultant who promises a position, a mention or a citation is promising something they do not control.
“Which of your case studies can I check?” Good answer: studies with a baseline, a time window, a named data source and a paragraph on what cannot be attributed to the work. Worrying answer: logos without numbers, or numbers without windows.
“What do you not do?” A consultant who can name the limits of their work has usually done it before.
What questions should you ask about access and process?
“What access do you need, and when?” Good answer: Search Console and analytics read access before the quote, so the proposal is based on your data rather than a guess. A crawl of the site is not the same thing as access to its reports.
“Who does the work?” If a senior person sells the engagement, ask whether that person does the analysis and attends the working sessions, or hands off to a team you have not met.
“How do you work with our developers and writers?” Good answer: a weekly working session, tickets in your tracker, and reviews of drafts before publication. Worrying answer: a monthly report and an invoice.
“What happens at the end?” Good answer: documents your team can act on without the consultant. Worrying answer: a dependency on their tooling or their retainer.
What should a good proposal contain?
- The symptom restated in the consultant’s words, so you can see they understood it.
- The order of work, with what each stage produces.
- The metrics, their sources and the baseline date.
- The access required and the date it is needed by.
- A fixed scope and price for the diagnostic phase, separate from any implementation phase.
- The things that are out of scope.
- A sentence on what the consultant will not promise.
A proposal that is mostly about the consultant’s history and tools, and only briefly about your site, is a brochure.
The checklist
Copy this into your notes and tick as you go.
- I have written down the symptom, who will implement, and the decision I want at the end.
- The consultant explained an order of work and why it is in that order.
- They named the metrics, the data source for each, and when the baseline is taken.
- They explained how they sample AI assistants and showed the prompt set or template.
- They said no when asked to promise a ranking, a mention or a citation.
- Their case studies state a baseline, a window, a source and the attribution limits.
- They asked for Search Console and analytics access before quoting.
- I know who does the work and who attends the sessions.
- The diagnostic is a fixed scope and price, separate from implementation.
- The proposal lists what is out of scope and what they will not promise.
How we answer our own checklist
In the interest of being checkable: our Diagnostic is a fixed-scope, fixed-price review across the six stages of the Crawl-to-Citation Method, with a prioritised plan and an AI-visibility baseline on a 40-prompt set. The Lab publishes the prompt set and the sampling method we use on our own site. Prices are on the engagements page. We do not promise rankings, mentions or citations, and the footer of every page says so.
What we don’t claim
- We do not claim this checklist identifies the best consultant for every situation. It filters out common failure modes; judgement still matters.
- We do not claim that a consultant who fails one item is unsuitable. Ask the follow-up question.
- We do not claim that AI-assistant visibility matters equally to every business. Measure your own AI-referral traffic first; our AI referral segment tool shows how.