Start by asking whether you need one at all
The most expensive mistake in this process is hiring well for a problem you do not have. Search takes months to compound, so if you need enquiries this quarter the honest answer is paid search or outbound, and an agency that takes your money without saying so is one to avoid.
There are other legitimate reasons to not hire. If nobody in your organisation can approve changes to the website, an agency will produce recommendations that sit in a backlog. If your category genuinely has little search demand, no amount of skill creates it. If your site has one severe technical fault, you may need a few days of development rather than a retainer.
A diagnosis costs you nothing with most reputable agencies and settles this before anyone quotes.
Ask who actually does the work
This is the single most useful question, and the answer is frequently different from the people in the room during the pitch.
Ask directly: who will write the content, who will make the technical changes, and will I speak to them. Some agencies sell through senior strategists and deliver through junior staff or subcontractors. That is not automatically wrong, and plenty of good work is delivered that way, but you should know before you sign rather than after.
Ask where the delivery team sits, too. Offshore and distributed teams are common and produce excellent work, including at this company, where the registered office is in London and the engineering team is in Lahore. What matters is that you are told plainly, that there is working hours overlap, and that you are not being quietly charged London rates for it.
Advice or implementation, and who is responsible for which
An audit that produces recommendations your developers never get to is the most common way SEO money is wasted. It looks like progress and changes nothing.
Establish which of three arrangements you are buying. The agency advises and your team implements, which works when you have developer capacity and a clear owner. The agency implements directly, which is faster and requires giving access. Or a hybrid, where they do the technical work and your team handles content.
All three are legitimate. The failure mode is leaving it undefined, because then nobody owns the backlog and six months later the audit findings are still open.
Promises that should end the conversation
A guaranteed position. Nobody controls Google's results. An agency promising first place for a term is either misunderstanding the work or expecting you not to check. The usual trick is guaranteeing rankings for phrases so specific that nobody searches them.
A fixed number of links per month. Links that can be bought in guaranteed monthly quantities are generally the kind that carry risk rather than value.
Secret or proprietary methods. There is no proprietary access to a search engine. Technique varies, judgement varies, access does not.
Immediate results. Technical fixes can move things within weeks because they remove an obstacle. Competitive commercial terms take months. Anyone collapsing that distinction is managing your expectations rather than informing them.
Ask what they would tell you not to do
This question is hard to fake and it separates judgement from capacity more reliably than anything else on this list.
Any real diagnosis produces findings that are not worth acting on: fixes that cost more than they return, keywords with volume but no commercial intent, pages better consolidated than improved. An agency that has looked properly will have opinions about what to skip.
An agency that proposes doing everything it found, in severity order, has either not prioritised or is being paid by volume. Both produce the same outcome for you.
Reporting: whose accounts, and what is being counted
Insist that Search Console, analytics and any rank tracking live in accounts you own and pay for. If the engagement ends you keep the history, the configuration and the data. Agencies that report exclusively from their own dashboard are, intentionally or not, making it expensive to leave.
Then look at what the report counts. Sessions, impressions and keyword counts all go up without enquiries going up. Agree before you start which single number both sides will judge the work by, and make it one connected to revenue.
Ask how they will handle a month where the number goes down, because it will. The answer tells you whether you will be given an explanation or a different metric.
Reading the proposal and the price
UK SEO is normally a monthly retainer, with one-off projects for migrations or technical work. Prices vary enormously and the headline figure is less informative than what sits under it.
A "from" price tells you the floor, not your price. Ask what the quoted figure includes in hours or deliverables, what falls outside it, and what triggers a change. Per-word content pricing is worth questioning specifically, because it rewards length and length is not what makes a page rank.
Check the contract term and the exit. Twelve-month lock-ins are common and not automatically unreasonable, since the work genuinely takes that long to compound, but you should know what happens to your accounts, your content and your data if you leave at month three.
Judging the evidence you are shown
Case studies are the standard currency and the easiest thing to dress up. A chart showing traffic rising tells you nothing about whether that traffic bought anything, or whether the client simultaneously launched a product.
More useful: ask for a client in your sector you can speak to directly, and ask that client what went wrong during the engagement and how it was handled. Every engagement has something. The answer tells you about the agency's behaviour under pressure, which is what you are really buying.
Be appropriately sceptical of statistics with no source. If an agency cites an average uplift figure, ask across how many clients, over what period, and how many engagements are excluded from the average.
A short list of questions worth asking
Who specifically will do the work, and can I speak to them before signing?
Will you implement the changes, or hand them to my team?
What did you find that you would recommend we do not bother fixing?
Whose accounts will the reporting come from, and what happens to them if we stop?
Which single number should we both judge this by in six months?
What would have to be true for you to tell us SEO is the wrong channel for us?
The last question is the most revealing. There is always an answer, and an agency that cannot produce one has not thought about where its work stops being useful.