CHOOSING & WORKING WITH AN AGENCY · September 2026 · ~10 min read
Questions to ask before hiring a marketing agency
The questions that matter are about who does the work, how it gets measured, and what happens when it does not work. The questions about tactics tell you almost nothing, because every agency has a rehearsed answer for those. Ask about the operating relationship instead.
On this page
- 01Which question tells me the most in the least time?
- 02What should I ask about measurement?
- 03How do I test a ranking factor claim in two minutes?
- 04What should I ask about their other clients?
- 05Which questions are a waste of time?
- 06How should they answer a question about AI search?
- 07What should I be asked in return?
- 08What to do this week
- 09When you do not need this
- 10Sources
- 11Related reading
- 12Questions about what to ask on your next call?
Most sales calls are structured so the agency talks for forty minutes and you talk for five. That is backwards, and it is why so many engagements start with a mismatch nobody noticed until month four.
You are not trying to catch anyone out. You are trying to find out how it will feel to work together in March, when the novelty is gone and someone has to tell you that a number went the wrong way.
01Which question tells me the most in the least time?
"Who will actually do the work, and can I speak with them today?"
The answer sorts agencies faster than anything else. Some will put the practitioner on the phone in ten minutes. Some will explain that the account manager coordinates a team you will never meet. Neither answer is disqualifying, but they describe two very different relationships and you should know which one you are buying.
Follow it with a second question. "What percentage of my work is done by someone outside your company?" Subcontracting is normal and often fine. Hidden subcontracting is the problem, because it means the person you evaluated is not the person producing. If they use partners, ask which parts and how they check the output, and understand the real tradeoffs between offshore teams, local shops, and individual freelancers before you decide it matters.
02What should I ask about measurement?
Ask what the number is today, before they touch anything.
If an agency cannot tell you your current calls, form fills, and map impressions before starting, they are going to report on activity instead of outcome. A number with no starting point is not a result, it is a screenshot. Every good engagement I have seen begins with someone writing down what things looked like on day one.
Then ask three follow ons. What will you report on monthly. What would tell us at ninety days that this is not working. And who owns the analytics and ad accounts if we stop.
That last one surprises people. It should be answered without hesitation, in your favor, in writing.
Ask also which of the deliverables they list are things you could verify yourself. A citation cleanup is verifiable in an afternoon. Ongoing optimization is not. That distinction is most of what separates a real scope from a padded one, and what a retainer should actually contain is a short and checkable list.
03How do I test a ranking factor claim in two minutes?
Worked example
Ask where the percentage came from, then do the division in front of them.
Local SEO pitches almost always contain a pie chart. Business profile signals are some share of the algorithm, reviews another share, links another. The chart looks like a measurement of Google. It is not.
Every version of that chart traces back to one survey. Whitespark's Local Search Ranking Factors report, published in its 2026 edition on 6 November 2025, asks 47 local search practitioners to spend about two hours scoring 187 factors from zero to five. The report was created by David Mihm in 2008 and has been run by Darren Shaw since 2017. It is the best thing of its kind and it is a poll.
Here is the arithmetic, and you can rerun it on the published table yourself.
Step one, find the ceiling. Forty seven respondents scoring a maximum of five gives a maximum possible combined score of 235. Primary business profile category, the top factor, scores 227, which is 96.6%derived of the ceiling.
Step two, measure the gaps. Proximity of address to the point of search scores 225. Keywords in the business title score 223. The distance between the first and third most important local ranking factors in the industry's reference document is four points out of 235, which is 1.7%derived of the scale.
Step three, ask how many people it takes to flip that. A single respondent moving one factor from a zero to a five changes a score by five points. One person out of forty seven, changing their mind about one factor, could reorder the top three.
That is not an attack on the report. It is what a survey is. The author says so himself, in a public comment on a Sterling Sky case study: the panel does not have access to Google's algorithm, much of their view is based on observation of what seems to help, and correlation is not causation.
So the debunk is specific: a share of the algorithm percentage is a number about a survey, not a number about Google. The group weighting percentages get quoted constantly, secondary summaries of the same report disagree with each other about them, and the chart they come from is rendered by script on the page rather than published as data. If a pitch tells you that profile signals are some exact percentage of local ranking, the honest follow up is: percentage of what, measured how, by whom, with how many people?
Where the survey is genuinely useful, and where it breaks down: it is a good instrument for deciding what to scrutinize and a bad instrument for deciding what caused something. There is at least one documented case where the panel was wrong and a controlled test proved it. Sterling Sky ran a test on review keywords, six people leaving reviews containing target phrases over several months at a business with no other SEO activity, and the rankings did not improve. In the first instance they got worse.
What should I ask about their other clients?
"Do you work with anyone I compete with, and how do you handle that?"
There is no single correct answer. Some agencies hold a hard exclusivity by city and category. Some work with several similar businesses in different neighborhoods and are open about it. What you want is a clear policy stated before you ask twice.
Then ask about capacity. "How many clients does the person on my account carry?" A practitioner running twenty accounts is not going to think hard about yours. A practitioner running four might be excellent, or might be new. The number itself is less interesting than whether they know it without checking.
Then ask the question most agencies do not publish an answer to, and hold us to it as well. How many clients did you have twelve months ago, how many do you have now, and how many left? SEOD publishes that it has served more than 100 clients and holds 128 verified five star reviews. Neither of those is a churn number, and a churn number is the more useful one. Ask every candidate for it, including this one, and treat "we do not track that" as an answer rather than a deflection.
Finally, ask for two clients who left and why. Anyone in business for a few years has some. An agency that claims none is either very new or not answering the question.
05Which questions are a waste of time?
"How many years have you been doing this?" Longevity is not skill, and this field changed materially in the last two.
"What is your process?" You will get a slide with five arrows on it. Ask instead what they would do first for you specifically, and why that before anything else.
"Can you get me to number one?" You are inviting a promise nobody can keep, and the agencies willing to make it are exactly the ones you do not want. There is a longer list of pitch language that should slow you down, and proposals tend to fail in a small number of recognizable ways.
Also skip the trivia. Asking whether your name, address, and phone number match across directories is a fine thing to care about, but the honest answer on consistency has become more nuanced than the checklist suggests, and the answer will not tell you whether these people are good.
06How should they answer a question about AI search?
With a number, a source, and a caveat. Anything else is atmosphere.
Ask what share of local searches actually produces an AI Overview. A candidate who has read the research will tell you that Ahrefs analysed 146 million search results pages in September 2025 and found AI Overviews on 20.5% of all of them, but on only 7.9% of local searches, and that Sterling Sky's independent tracking sees AI powered local packs on roughly 7% of tracked keywords.
A better candidate will volunteer the contradiction. A Whitespark study of 540 manual queries across six local industries reported AI Overviews on a far higher share of local business queries, because it deliberately used long conversational phrasing rather than the short "service plus city" pattern that dominates a large keyword database. Both samples are right about themselves. Short transactional local searches are still local pack territory. Long "help me choose" local searches are AI territory. An answer that gives you one number without the other is selling you a direction.
The way people ask AI assistants for a nearby business is not the same as a search query, and a real answer sounds concrete rather than futuristic.
07What should I be asked in return?
Watch what they ask you. It is the strongest signal in the meeting.
An agency that is going to do good work will ask what your margins look like by service, which jobs you want more of and which you would rather stop taking, how leads reach you today, and who answers the phone at 6pm. Those questions come from someone planning to make you money rather than traffic.
An agency that asks only for your website and your budget is going to sell you a package.
Ask them directly: "What do you need from me every month, in hours?" Any engagement requires approvals, photos, answers, and access. If they say none, they are either doing something generic or they will stall in week three waiting on you.
08What to do this week
Write down five questions before your next call and refuse to be moved off them.
Mine would be these. Who does the work. What are my numbers today. What would make you tell me at ninety days that this is not working. Who owns the accounts if we part. How many clients left in the last year.
Send them in advance. An agency that prepares answers is showing you how they will handle a Tuesday request later. One that improvises through all five is showing you that too.
Then take any percentage in their deck and run the two minute test above. Ask what the denominator is and how many people are inside it. You are not being adversarial, you are asking a vendor to explain their own slide, and the quality of that explanation is the most information you will get for free all week.
Be honest with yourself
When you do not need this
If you are hiring for one defined project, a site rebuild or a listing cleanup, most of this is overkill. Scope it, agree a price, check the output, done. Interrogating a vendor about ninety day measurement makes no sense for a two week job.
If you already have a marketer inhouse who is good and simply underresourced, you may need a contractor for specific tasks rather than an agency relationship. Different purchase, different questions.
And if you are choosing between two agencies you already trust from direct experience, stop researching. You are optimizing a decision that is already good enough.
Sources
- Whitespark, "Local Search Ranking Factors," Darren Shaw, 6 November 2025. Forty seven experts, 187 factors, two hour survey. Source of the scores used in the arithmetic above. Vendor research, and expert opinion rather than test data. Whitespark sells local SEO tools.
- Sterling Sky, Joy Hawkins, "Can Keyword Text from a Google Review Boost Your Local Map Pack Rankings?". The controlled test where the panel view did not hold, and the comment thread containing Darren Shaw's own description of the survey as opinion.
- Ahrefs, "What Triggers AI Overviews? 86 Factors and 146 Million SERPs Analyzed," Ryan Law. Desktop results pages, September 2025. Source of the 20.5% overall and 7.9% local trigger rates. Vendor research: Ahrefs sells SEO software.
- Sterling Sky, "The State of Local SEO in 2026," Joy Hawkins. Independent tracking of AI powered local packs at roughly 7% of keywords, using third party rank tracking data. Vendor research from an agency that sells local SEO.
Related reading
- The claims agencies make that do not survive checking. The five ways a statistic in a pitch fails, with the traces.
- What a good discovery process looks like. What should happen after you have chosen, and how to tell it is happening properly.
- Setting expectations for the first 90 days. The timeline conversation these questions are supposed to produce.
- Reference checks that actually tell you something. How to verify the answers you just collected with someone who has lived them.
Questions about what to ask on your next call?
Email me at eric@seod.com with your industry and the type of agency you are talking to, and I will send back the five questions most likely to expose a bad fit in your specific category. Dental and home services fail in different places, and the questions should be different too. No pitch attached, and no sequence afterward.
Otherwise there is more on hiring and working with agencies here.