How to choose an SEO agency, and which ones to avoid

The two things buyers compare are price and the list of deliverables, and neither one discriminates. Two agencies quoting the same number can be doing opposite work, and a deliverables list of four articles and twelve links describes activity rather than outcome. What actually separates them is the questions they ask you before quoting anything. A provider who has not asked to see your Search Console data is proposing work without having looked at the demand, and that is guessing regardless of what the proposal costs.

How do you choose an SEO agency?

By how it diagnoses before it proposes, not by what it promises. A new business needs someone willing to say wait six months; a mid-authority site needs someone who prioritises from data that already exists; a consolidated site needs someone who knows what not to touch. Anyone guaranteeing position one rules themselves out.

Quick answer

• Judge by the questions they ask you, not by the deliverables they list.
• Nobody serious quotes without seeing your data. Search Console access comes before the proposal.
• Any promise of a specific position or timeline ends the conversation. It is not demonstrable.
• A report without a fixed set of queries cannot be compared to last month's.

What kind of decision this is

The mental model that makes this hard is that evaluating an SEO proposal requires understanding SEO. Under that assumption the only defence is getting three quotes and picking the middle one, which is how most of these decisions actually get made.

The way out is to notice that you are not evaluating the answers, you are evaluating the process that produced them. A proposal written after looking at your data reads differently from one written from your homepage, and the difference is visible without technical knowledge: it references your queries, your pages, your specific problems.

There is a fairness point too. Most disappointing SEO is not fraud. It is capable people measuring badly, or producing content nobody asked them to prioritize because nobody agreed on priorities. The questions below are designed to surface that, not to catch criminals.

The five questions to ask

1. How many hours a month, and who does them?

SEO is almost entirely skilled time. There is no raw material and no inventory, so when a price drops the hours drop with it, or the seniority does.

A provider who cannot answer this is selling a package. The answer also lets you do arithmetic: a 500-dollar retainer cannot contain twenty hours of experienced work.

2. What happens in month one?

The correct answer usually involves auditing and produces nothing visible. That is a good sign, not a bad one.

Be wary of the opposite: a provider who starts publishing in week one has not looked at your data, which means the content is a guess about what you need. The order matters, and it is covered in the Spanish guide on technical SEO.

3. How will this be measured?

If the answer is "rankings", ask a follow-up: rankings for which queries, and will that list stay the same every month?

A report built on a rotating set of queries cannot show progress, because each month describes something different. A fixed cohort can. This single question separates measurement from decoration, and the reasoning is in the Spanish guide on reading Search Console.

4. What happens if six months pass with no result?

An honest answer exists, and it is not "that does not happen". It sounds like: we would re-examine the diagnosis, here is what we would look at, and here is the point at which we would tell you this channel is not the right one for your business.

A provider who cannot imagine failing has not thought about the mechanism.

5. Can you show me something you decided not to do?

The best question on this list, and the one nobody expects.

Deciding not to pursue a query means having measured it, compared cost against return, and accepted that some available work is not worth doing. A provider with a good answer here has just demonstrated the entire craft in two minutes.

The signals that end the conversation

A guaranteed position or timeline. Not demonstrable. Nobody controls the algorithm, and 2026 alone had four confirmed updates, with the March one changing position for 79.5 % of the top three.

A proposal that arrived without any questions. It is a template with your logo on it.

Reporting that only shows average position. It mixes queries of different intent into one number that says almost nothing.

Refusal to explain the first month. Either there is no plan or the plan is to start producing immediately.

Ownership access to your Google properties as a condition. Restricted access is enough to read the data. Handing over ownership is easy and unpleasant to reverse.

When not to hire anyone

When nobody searches for what you sell. If the queries do not exist, there is no demand to capture, and the money works harder creating demand elsewhere.

When the site does not convert. Sending traffic to a page that does not sell is paying to confirm a problem you already had.

When you need revenue this quarter. SEO does not arrive in time, and pushing it produces exactly the timeline promises listed above.

When an independent audit would answer the question first. A bounded diagnosis costs a fraction of a retainer and sometimes concludes that the retainer is unnecessary.

What the bands buy

The full breakdown, including why serious retainers ask for six months, is in how much SEO costs for a small business.

Every guide lists their red flags. None lists yours.

Read eight guides on this and you will get the same list: guaranteed rankings, a proposal sent before any questions were asked, vague references to "the team" with no names, link building that sounds fast or automated. That list is correct and it is worth using.

What none of them contains is the other half, and in my experience it decides more outcomes: the conditions on your side that make any agency fail.

Nobody who can answer business questions. What a customer is worth, which services carry the margin, why last year was different. An agency that cannot get those answers writes generic content, and then gets blamed for writing generic content.

Nobody who can approve anything. Drafts that sit for three weeks turn a six-month engagement into a two-month one, at the same price.

No agreed definition of working. If nobody wrote down what counts as success before signing, the argument in month five is unwinnable for both sides.

And an unanswered phone. In service businesses this ends more engagements than any ranking issue, and no provider can fix it from outside.

One statistic circulating on this topic claims nearly 75 % of agency switches come from expectations misaligned before signing. No source is named for it anywhere I could find, so treat it as a plausible framing rather than a measurement. The direction matches what the four conditions above describe.

Three questions to answer before calling anyone

What the work will cost and for how long, which comes from the value of a customer rather than from a price list: SEO cost for a small business.

What a serious proposal should contain, so you can tell a diagnosis from a package: what an SEO proposal should include.

And how the result will be judged, which is agreed before signing and not after: how to set up Search Console and Core Web Vitals in Search Console.

If there is already a contract running that is not producing, the diagnosis is a different one: is my SEO agency working. And if the business serves customers at an address, most of the work is somewhere else entirely: local SEO for small business.

That address also changes who is worth hiring, because three different businesses sell local work under the same title and one of them is reselling somebody else's hours. Which figure does what, and the questions that separate them, are in what a local SEO expert actually is.

Mistakes that repeat

• Comparing quotes by the number. The same figure covers opposite work.
• Accepting a proposal written without your data. It is a template.
• Signing for three months. It ends right before the work becomes measurable.
• Granting ownership access. Restricted access reads the same data and reverses cleanly.
• Asking for rankings instead of decisions. One is decoration, the other is the craft.
• Skipping the independent audit. It sometimes concludes you do not need the retainer.

Data and transparency

The red-flag list (guaranteed rankings, proposals sent before questions, unnamed teams, vague link building) reflects the consensus of the first results for this query on 12 August 2026. The figure of nearly 75 % of agency switches caused by misaligned expectations is quoted as unsourced: none of the pages carrying it names a study. The four client-side conditions are operating judgement from work on a portfolio recording more than 300 million annual impressions in Search Console.

The market ranges of 300 to 2,500 USD a month and 500 to 5,000 USD per one-off project come from SE Ranking's SEO pricing survey published 13 December 2024, covering 260 agencies across North America, Europe and Oceania, whose authors note the sample may not reach statistical significance; in it 64 % of agencies charge under 1,000 USD a month. The 2026 algorithm update data, including the March update changing position for 79.5 % of the top three, comes from SE Ranking. No statistics on agency performance or client satisfaction appear here: figures of that kind circulate widely and none I found could be traced to a primary source. The questions listed come from operational experience across a portfolio recording more than 300 million impressions a year in Search Console. I sell these services, which is disclosed here rather than woven into the argument. Verified as of August 2026.

Primary sources, opened on 13 August 2026: the SE Ranking survey.

What this changes

There is a test hidden in this list that applies to the buyer as much as the provider.

Every question above assumes someone will ask it and then actually read the answer. A buyer who never asks for a fixed-query report will keep receiving rotating ones, because rotating ones are easier to produce and nobody objected. The engagements that quietly go wrong are rarely the ones where a provider hid something. They are the ones where nobody asked a question that had a wrong answer.

Frequently asked questions