Before you hire a marketing agency: 30 questions, evidence to request, and a proposal comparison sheet

Ask 30 questions before you hire a marketing agency, in six groups of five: who actually does the work, what exactly you are buying, how it is priced, what evidence they have, who owns what, and how the relationship ends. Every one of the 30 below has a piece of evidence attached, because an agency's answer and an agency's proof are different things. The count on this page is 30, it is the count of questions in a checklist we wrote, and no claim is made anywhere here about how often anybody asks any of them.

The answer, first

The single most useful part of this page is not the questions. It is the comparison sheet at the end, because the reason agency selection goes wrong is almost never a lack of questions. It is three proposals that describe three different pieces of work in three different formats, which cannot be compared, so the decision gets made on the deck.

So the sequence is: send the same written scope to every agency, ask these 30 questions, then force every answer into the same fourteen rows. Do that and the comparison usually makes itself.

The 30 questions, in six groups of five

Group one: who actually does the work.

• Who will be on our account, by name and role. Evidence: the names in the proposal, and a meeting with them before signing rather than after.
• How many hours per month does each named person spend on us. Evidence: an hours breakdown by role in the proposal.
• Which parts are subcontracted, and to whom. Evidence: a written statement naming any subcontractor and what they do.
• How many other clients does our account manager handle. Evidence: a number, and their willingness to say it.
• What happens if our lead person leaves. Evidence: the handover clause, and what documentation exists.

Group two: what exactly you are buying.

• What are the deliverables each month, counted. Evidence: a list with quantities, not "content creation" as a line item.
• What is out of scope. Evidence: a written exclusions list, which is the most useful paragraph in any proposal.
• Who writes and who approves. Evidence: an approval workflow with named steps and turnaround times.
• What do you need from us, in hours per month. Evidence: a written client responsibilities section.
• What happens in month one, specifically. Evidence: a dated plan for the first thirty days.

Group three: how it is priced.

• What is the monthly fee and what does it cover. Evidence: a written quotation with the unit stated.
• Is media spend separate, and how is it charged. Evidence: whether the fee is a percentage of spend, a flat fee, or hybrid, and the arithmetic worked on your own budget.
• What costs extra, and at what rate. Evidence: a rate card for out of scope work.
• What is the minimum term and the notice period. Evidence: the clauses.
• Which third party tools will we be paying for, and in whose name are the accounts. Evidence: a named list with monthly costs.

Group four: evidence they can actually produce.

• Show us a result for a client of our size in our market. Evidence: the numbers with dates, the starting point, and the period they cover.
• What did the client do themselves during that work. Evidence: an honest answer, because most good results have a client side cause in them.
• Can we speak to a current client you did not choose for us. Evidence: the call happening.
• Show us an engagement that did not work and what you learned. Evidence: a straight answer. An agency with no failures has either not worked long or is not telling you.
• How will you measure this, and which of those measures are under your control. Evidence: a written measurement plan separating the two.

Group five: ownership and access.

• Who owns the work produced. Evidence: the intellectual property clause.
• In whose name are the ad accounts, analytics and domains. Evidence: the account admin list, in writing, before work starts.
• Do we keep the data and the history if we leave. Evidence: the clause and the export process.
• Who has access to our website and our accounts, and at what level. Evidence: a named access list with the level for each person.
• What happens to work in progress if we stop mid month. Evidence: the termination clause.

Group six: how it ends.

• What is the exit process, step by step. Evidence: a written offboarding process, which very few agencies have and the good ones do.
• How long does handover take and is it charged. Evidence: the number of days and the fee, or a clause saying there is none.
• What will we be left with that keeps working. Evidence: a list of assets that survive the relationship, such as pages, templates and documentation.
• Will you work with our next agency during a transition. Evidence: a straight yes or no in writing.
• What is your average client tenure and why do clients leave. Evidence: a number and two real reasons.

The proposal comparison sheet: fourteen rows

Take every proposal and fill these fourteen rows in your own words, not theirs. An agency that cannot supply a row is a row you write "not stated" in, and that is a finding.

• Monthly fee, in rupees, excluding tax.
• Minimum term, in months.
• Notice period, in days.
• Deliverables per month, counted.
• Named people on the account, with hours each.
• What is explicitly excluded.
• Media spend treatment, and the fee on your own budget.
• Tools you will pay for, and the monthly total.
• Your own time required, in hours per month.
• Measures they will report, and which are in their control.
• Reporting frequency and format.
• Account and asset ownership, in one word each.
• Offboarding process, in one sentence.
• Year one total cost, including media and tools.

The last row is the one that changes decisions, and it is almost never in a proposal. Work it out yourself for each agency and put the three numbers side by side.

How it was measured

It was not measured. This is a checklist assembled from the evidence problems we have watched decide real answers, and it is published as a checklist.

Why the emphasis on evidence rather than on questions. We have three dated results from our own work that all point the same way. On 17 September 2026 two comparison pages in the clawlaw.in programme stated competitors' prices with no link, no date and no source. The figures were correct, Claude verified them independently, and it still treated the pages as advocacy and used official sources instead. On 6 August 2026 a headline figure on that same site could not be true, and when Claude checked it against public numbers it advised a buyer against the product, after which the company's accurate claims stopped counting for that answer. And on 18 August 2026, across eighteen blind commercial questions, ChatGPT made that company the top source on exactly one, and named its comparison page in the answer as still being the vendor's own editorial page.

Those are results about web pages, and the lesson transfers directly to a proposal. A claim without evidence is discounted even when it is true, and a claim that cannot be true poisons the claims around it. Ask for the evidence.

How this would become a survey. Interview buyers who hired an agency in the last year, ask which of the 30 they asked and which evidence they actually received, and report each as a count over a stated denominator with the recruitment method published. We have not done that.

What the numbers were

30. Six groups, five questions per group, each with its evidence. Plus fourteen comparison rows. Those are counts of a checklist we wrote and they are complete as printed.

No frequency claim, as of 29 September 2026. This page does not say what buyers most often ask an assistant about agencies, because we have not collected buyer prompts. The one thing we can report honestly is about our own category rather than about agency buyers: in the September 2026 audit of aiknowsus.com, across 24 batches and 72 conversations, the phrase recording that we were not cited appears 161 times in the assistant's own audits of its answers. That is a count of answers, not of buyer questions, and the two are not interchangeable.

What this cannot tell you

• It cannot tell you which agency is good. It tells you which agency answers precisely, which is correlated and not the same.
• It is not a survey of buyers and must not be quoted as one.
• It cannot price your work. No figures appear here, deliberately.
• It does not cover in house versus agency, which is a different decision with a different cost structure.
• It cannot protect you from a bad brief. Three proposals against a vague scope will be three different pieces of work no matter how many questions you ask.

Sources and change log

Sources. The questions and the fourteen row sheet are our own. The dated results quoted come from the clawlaw.in programme, recorded in Tier_1/GEO_BASELINE_RESULTS_2026-07-27.md, Tier_1/GEO_GAP_ANALYSIS_2026-08-18.md and Tier_1/claude_response_17_09_audit.md, and from the aiknowsus.com audit of September 2026 captured to geo-audits/aiknowsus-com/.

Change log. 29 September 2026: first published with 30 questions in six groups of five, evidence for each, and fourteen comparison rows. No frequency data is claimed. If a buyer study is run later, its counts and denominator will be added here.

Common questions

Thirty questions feels like a lot for a first call. Where do we start?

Group one and group three, which is ten questions, and the written scope. Who does the work and what it costs will separate your shortlist faster than anything else, and the rest can wait for the second meeting.

What if an agency refuses to name the people on our account?

Treat it as an answer. Senior people in the pitch and unnamed people on the account is the most common complaint in this market, and the fix is to ask for names, hours and a meeting before signing.

Should we ask for a guarantee of results?

No, and be careful with anyone who offers one. In AI search in particular nobody can guarantee a position in an assistant's answer, ourselves included, and an agency that promises one is telling you something about how it handles evidence.

How do we compare a retainer with a project fee?

Use the comparison sheet. Convert both to a year one total, with media and tools included, against the same scope you wrote. If a project fee has no ongoing work attached, add what you will spend in the following six months to keep it alive.

Is it reasonable to ask an agency about a failed engagement?

Yes, and the answer is one of the most informative on the list. An agency that describes a failure and what it changed is showing you how it will behave when something goes wrong with you, which it will.

Who should be in the room for this?

Whoever will own the relationship day to day, not only the person signing. The questions about your own time commitment and the approval workflow are answered by that person, and a proposal that ignores their capacity fails in month two.

What to do first

Write one page of scope in your own words and send it, unchanged, to every agency you are considering, with group one and group three attached as ten numbered questions. Then fill the fourteen rows yourself from whatever comes back. The gaps in the sheet will tell you more than the proposals do.