The hardest thing about hiring someone to do SEO is that if you could evaluate their work, you would not need to hire them. You are being asked to judge expertise you do not have, in a field with no licensing, no audited performance numbers, no agreed standards, and a strong tradition of confident storytelling. It is buying a mechanic's opinion when you cannot open the bonnet.
I have sat on both sides of this. I have been hired, I have been passed over for people who were worse and better at selling, and I have been asked by boards and by founders to sit in on the evaluation of somebody else. What follows is what I actually use.
One principle underneath all of it: do not test knowledge. Test judgment. Knowledge is free now. Anyone can produce a fluent explanation of canonicalisation or crawl budget in ten seconds, and the fluency tells you nothing. What separates people who can do this work is knowing what to do when the answer is unclear, what not to bother with, and when to say the uncomfortable thing. That is what you have to probe.
The Six Questions
1. "Tell me about a time SEO was the wrong investment."
The single most useful question I know. You are asking someone whose income depends on selling SEO to describe a situation where they should not have been hired.
A good answer is specific and slightly painful. A brand-new marketplace with no inventory on either side, where organic could not do anything until the supply problem was solved. A B2B company with eleven possible customers on earth, where the entire addressable search volume was smaller than one salesperson's call list. A business whose economics needed payback in six weeks. Good answers usually end with "so we told them to spend it on something else" or "we should have, and we didn't, and it went badly."
A bad answer treats the question as a trick. "SEO is always worth it long term." "It is the highest-ROI channel there is." Anyone who cannot describe the boundaries of their own discipline either has not worked at the boundary or will not tell you when you are at it. That is exactly the person who will keep billing you through a year when the money should be going elsewhere.
2. "What would make you tell me this site cannot rank for our main term?"
This tests whether they can assess difficulty honestly rather than sell hope.
Good answers talk about who currently occupies the results and why. Whether the page one incumbents are structurally different from you: aggregators, marketplaces, institutions, publishers with twenty years of accumulated authority. Whether the query intent even matches what you sell, which is the failure I see most often and the one clients resist hardest. Whether the SERP is dominated by features that push organic results below the fold. Whether the gap is closeable with content and links, or whether it needs a different business, which is sometimes the true answer.
Bad answers reach immediately for a single difficulty score from a tool, or say that anything is rankable with enough quality content, which is the astrology of this industry.
3. "Walk me through the first two hours of a forty percent traffic drop."
This is the closest thing to a practical exam. You are listening for a diagnostic order, not a list of causes.
Good answers start by defining the drop before explaining it. Which segment, which pages, which queries, which devices, which countries. Is it impressions or clicks or both, because those point in completely different directions. Is it outside normal variance for this business, or is this a Tuesday being compared to a Sunday. Only then do they move to causes, and a good practitioner checks self-inflicted causes first: a deploy, a robots or noindex change, a CDN or firewall rule now blocking crawlers, a certificate problem, a migration finishing its recrawl. Algorithm updates come late in a good answer, not early.
Bad answers lead with "there was probably a core update." That is the horoscope. It explains everything, predicts nothing, and conveniently locates the cause outside the vendor's control.
4. "What is a technical fix you implemented yourself, in the code?"
You are separating people who identify problems from people who solve them. Both are legitimate, but you need to know which you are buying, because a purely advisory hire plus an engineering team that has no capacity equals a very expensive stack of tickets nobody actions.
Good answers get concrete fast and usually involve a specific mess: fixing a redirect chain in a config file, writing a script to find orphaned pages in a crawl export, correcting a template that emitted the same canonical on every page, building something small to compare a sitemap against what actually returns 200. The specificity is the signal.
A perfectly acceptable answer is "I don't code, I write the spec and work with engineers." That is honest and workable. What should worry you is vagueness in both directions: someone who implies deep technical capability but cannot name one thing they actually changed.
5. "What do you not know?"
Ask it plainly. The reaction in the first two seconds is most of the information.
Right now, in 2026, anyone honest has a long list, and near the top of it should be how answer engines actually select and cite sources. I have argued at length that nobody has reliable data on this yet. Someone who claims a proven methodology for getting cited by language models is telling you something important about their relationship with evidence.
Good answers name real gaps and describe how they work around them: running their own tests, watching a metric they trust, being explicit about what is a hypothesis. Bad answers produce a humble-brag ("I'm always learning, the field moves so fast") or, worse, cannot think of anything.
6. "What would you need from us, and what happens if you don't get it?"
Anyone who has actually delivered results knows the work fails on the client side more often than the vendor side. They will have a list: engineering time, access to analytics and Search Console, someone with authority to approve content, subject-matter experts who will answer questions, permission to change page templates.
The good version includes a consequence. "If I cannot get roughly a day of engineering time a month, the technical recommendations will sit in a backlog and you should not expect the technical portion of the forecast." That is a person who has been burned and has learned to say so before signing rather than after.
Somebody who says they need nothing from you is either doing something superficial or has not thought about it.
Red Flags
Guarantees. Guaranteed rankings, guaranteed traffic, guaranteed timeline. Nobody controls the ranking system. A guarantee is either meaningless (guaranteed page one for your own brand name) or a lie.
A proprietary framework with a trademark. Usually the standard work with a name on it, priced for the name. Occasionally there is genuine method underneath, so ask what it would tell them to do differently from someone without it. Watch whether the answer is specific.
Domain authority as a primary metric. It is a third-party tool's invented score, not something search engines use. Optimising for it is optimising for a vendor's model of reality.
Deliverables counted by volume. Twenty blog posts a month, fifty links a quarter. Volume commitments describe an output the vendor can control regardless of whether it works. It is a manufacturing contract for content, and it produces exactly what you would expect.
Immediate condemnation of the previous agency. Some predecessors were genuinely bad. But a candidate who leads with how catastrophic the last team was, before doing any analysis, is running a sales motion. The reflex to explain the last year's numbers by pointing at somebody who has left the room does not stop once they are hired.
An audit as the pitch. A free audit given before signing is a sales document, not an analysis. I have written about this. It is generated from a crawler, sorted by severity, and designed to alarm. The tell is that it lists everything and prioritises nothing.
Nobody ever says no. If every idea you float in the meeting is a great idea, you are talking to an order-taker. You are paying for the "no."
The Cheap Test That Beats the Interview
Interviews reward people who interview well. If the engagement is worth real money, buy a small piece of work first.
Pay two or three candidates for a bounded piece of analysis. Give them the same brief, the same access, two weeks, and a fee that respects their time. Something like: here is our site and our Search Console access. Tell us the three things you would do first, why those three and not others, what you would expect to happen, and how we would know if you were wrong.
This costs a fraction of a year's retainer and it is the most informative thing you can do. Reading three responses to an identical brief is enormously clarifying even if you know nothing about SEO, because you are no longer judging expertise in the abstract. You are comparing.
What to look for: Did they prioritise, or did they list? Did the recommendations come from your actual data or could they have been written about any company? Did anyone tell you something you did not want to hear? Did they say what they expected to happen, with a number and a timeframe, and how you would know if it failed? Did anyone ask you a question about the business rather than only about the website?
That last one matters more than it looks. The best diagnostic question I have ever been asked by a candidate was about margin, because it changed which pages were worth ranking. Someone who never asks how you make money is going to optimise for traffic, and traffic is not the thing you needed.
After You Hire
Agree, in writing, in the first two weeks: what we expect to move, roughly by when, and what we will conclude if it does not. Not a guarantee, a hypothesis. The point is that six months later there is a written record of what was predicted, so the conversation is about evidence rather than vibes.
Insist that reporting separates branded from non-branded organic traffic. If it does not, you cannot tell whether SEO is working or whether your paid campaigns and PR are driving people to search your name. An enormous number of SEO success stories are brand campaigns with a different invoice on them.
Expect the honest cadence: very little for a few months, then compounding. Anyone showing a straight line up from week two is showing you a metric chosen for its shape.
And keep noticing who tells you inconvenient things. Over a year, that is a better signal than any credential, any case study, and certainly any audit.
Hire the one who told you something you did not want to hear, early, when it would have been easier not to.