Signs your SEO agency is not doing anything

You are six months into a retainer, the traffic looks flat, and you have no way to tell whether that means the work is slow or the work is absent. That is the honest position most business owners are in, and it is not a knowledge problem. You cannot evaluate the work, but you can evaluate the traces the work leaves, and every check below can be done by someone who has never read a word about SEO. Two of them take under a minute.

How do you know if your SEO agency is doing anything?

By checking whether impressions rise while clicks do not. On a new site six flat months can be normal; on a mid-authority site a flat year is not; on a consolidated one, holding ground is already a result. The question is not whether they work, it is what explains the number.

Quick answer

• Look for traces, not for expertise. New URLs, changed titles, documented decisions.
• If nobody ever asked for Search Console access, they are working blind.
• A report that shows a different set of queries each month cannot be compared to the previous one.
• Some signals that look like inaction are good work. A first month with nothing published is one.

What kind of problem this is

The assumption that traps people is that judging this requires understanding SEO. Under that belief the only options are trusting the provider or hiring a second one to audit the first, and both are expensive.

The way out is narrower. Judging whether the decisions were correct is not the task. The task is establishing whether decisions were made at all, and that leaves evidence a non-specialist can see: files that changed, titles that differ, reports whose structure stays comparable month to month.

There is also a fairness point worth making early. Most bad SEO is not fraud. It is competent people measuring badly, and the checks below distinguish absence of work from absence of results, which are different accusations.

The five checks

1. Are there new URLs in the last ninety days?

Open the sitemap, usually at /sitemap.xml, or the pages report in Search Console. Sort by date if you can.

If nothing new exists and nobody explained why, that is a straightforward finding. If new pages exist but nobody told you they were coming, that is a communication problem rather than a work problem.

2. Did any existing title change?

Titles are cheap to change and often the highest-return work on a site. In an audit of a 312-page site, the single most profitable action in the whole project was rewriting titles: one page held position 1 with zero clicks because its title did not contain the term people typed.

If six months have passed and not one title on the site is different, the highest-return lever has not been touched.

3. Does the monthly report show the same queries each time?

This one takes five minutes and catches more bad work than the other four combined.

Open two reports from different months and compare the query lists. If the set changes every month, no comparison is possible, because you are looking at different things each time. A report built on a fixed cohort of queries can show movement; one that re-picks its queries monthly can show whatever it likes.

4. Did anyone ever ask for Search Console access?

Not analytics. Search Console specifically, which is where impressions, positions and click-through rate live.

A provider working without it is proposing content and technical fixes without seeing which queries the site already appears for. It is possible to do, and it is guessing. The reasoning behind that data is in the Spanish guide on reading Search Console.

5. Can they name a decision, not a deliverable?

Ask what they decided not to do this quarter, and why.

A good answer describes something measured, compared and rejected. A weak one lists outputs: four articles, twelve links, one report. Counting deliverables is how work gets measured when nobody is measuring outcomes.

When the signal does not mean what it looks like

This is the section that makes the rest usable, because several signs of good work look exactly like inaction.

A first month with nothing published. Auditing is the correct use of month one, and it produces nothing visible. A provider who publishes in week one without having looked at your data is producing blind, which is worse.

Technical fixes you cannot see. Removing a stray noindex, fixing redirect chains, restructuring internal links: none of it changes how the site looks. Ask for the list; do not expect to notice it.

Rankings that do not move in a straight line. Positions oscillate, and Google confirmed four algorithm updates in 2026, with the March one changing position for 79.5 % of the top three. A flat quarter that coincides with an update is not evidence of anything.

Traffic flat while conversions rise. If the queries changed toward buying intent, that is the outcome you wanted and the traffic chart will not show it.

When the honest answer is to stop

When no amount of checking resolves it. If the provider cannot produce a fixed-cohort report and cannot name a decision, the relationship has no measurable content, and paying longer will not create one.

When results genuinely have not arrived after a full cycle. Serious engagements run six months for reasons of sequence, but six months is also long enough to expect evidence of work even where results lag.

Three explanations for "no results", and how to tell them apart

Every article on this lists signs that an agency is underperforming. None points out that the same flat graph has three very different explanations, and confusing them costs either an unnecessary switch or another year paying for something that will not work.

One: the work is not being done. You recognise this by looking at the site rather than the report. If six months produced no new pages, no rewritten copy and no changes you can point at, the report is beside the point.

Two: the work is being done and the plan was wrong. The expensive one, because it looks productive. New content every month, rankings improving, all of it on queries nobody uses when they are ready to buy.

Three: the work is being done, the plan is right, and the market moved. This one is recent and almost nobody includes it. Zero-click searches went from 56 % to 69 % between May 2024 and May 2025, according to Similarweb, with the drop concentrated in informational queries. A site can hold its rankings, grow impressions and lose clicks without anyone having done anything wrong.

Separating them takes the same check in all three cases and needs no expertise: look at impressions and clicks separately, month by month, over the same set of queries. If impressions rise and clicks do not, it is not the first case. How to read that without fooling yourself is in how to set up Search Console, and what to do about the divergence itself is in answer engine optimization vs SEO.

What to check before deciding anything

Whether the site changed at all, which is a five-minute look rather than a report reading.

Whether what improved has commercial intent, because rankings on questions nobody asks before buying are a real result on the wrong target: free keyword research.

Whether the local layer was even touched, if the business serves customers at an address, because that is usually where the return was: local SEO for small business and Google Business Profile.

And what was agreed as success before signing, which is the question that decides the argument and almost never got written down: how to choose an SEO agency and what an SEO proposal should include.

If the conclusion is that the plan was never right rather than that the work was never done, the starting point for the next one is what is SEO.

Mistakes that repeat

• Judging by traffic alone. It moves for reasons that have nothing to do with the provider.
• Accepting reports with a rotating query set. Nothing in them can be compared.
• Expecting visible output in month one. Auditing produces nothing and is the right use of it.
• Confusing absence of results with absence of work. They need different conversations.
• Firing during an algorithm update. The next provider will start by redoing the diagnosis.
• Asking for rankings instead of decisions. Average position is the metric that says least.

Data and transparency

That zero-click searches went from 56 % to 69 % between May 2024 and May 2025 comes from Similarweb's Generative AI Landscape 2026, and that the drop concentrates in informational queries comes from Seer Interactive, measured June 2024 to September 2025. No figure appears here for how often each of the three causes applies: no measurement is published and they are separated case by case.

The 312-page site with a position-1 page receiving zero clicks is my own audit work; it is client work, so the site is not named and the figures are rounded. The 2026 algorithm update data, including the March update changing position for 79.5 % of the top three, comes from SE Ranking. The checks described here are operational judgment from work across a portfolio recording more than 300 million impressions a year in Search Console. No statistics on how often agencies underperform appear in this article: figures of that kind circulate widely and none that I found could be traced to a primary source. Verified as of August 2026.

Primary sources, opened on 13 August 2026: the Seer Interactive study; the SE Ranking survey.

What this changes

The uncomfortable part of this list is that it tests the provider and the client at the same time.

Every check above assumes someone will ask for something specific and then read the answer. A client who never asks for a fixed-cohort report will keep receiving rotating ones, because rotating ones are easier to produce and nobody complained. The relationships that go bad quietly are rarely the ones where the provider was hiding something. They are the ones where nobody asked a question that had a wrong answer.

Frequently asked questions