Somewhere around month five the tone of the monthly call changes. The reporting still arrives, the activity is real, and nobody can point to anything that has actually moved. The agency talks about foundations being laid. You start wondering whether you were sold something that does not work, or whether you hired the wrong people, or whether the whole category is overstated.
All three are possible and they need different responses. Before firing anyone or writing off the channel, work out which situation you are actually in, because the most common answer is none of the three. Switching AI SEO Agency partners solves roughly a third of stalled engagements and wastes another two months in the rest.
First, Separate the Four Things That Could Be Wrong
They look identical from the client seat and have completely different fixes.
The work was blocked. Something technical prevented any of it from taking effect. Crawler access, rendering, a staging environment that never went live, schema that was specified but never deployed. Everything downstream produces nothing when this is true, and it is more common than agencies like to admit because it is usually discovered late.
The work targeted the wrong things. The activity happened correctly against a question set nobody asks. This is the most common failure and it traces back to whether the question list came from sales conversations or from a search volume export.
The work was too thin to register. Right target, insufficient depth. Frequently a scope problem rather than a competence problem, and the honest version is that the budget bought less than the category requires.
The work is fine and the timeline was misrepresented. Large content estates take longer. If you have thousands of pages and were sold a two month result, month five with early signals may be exactly on track.
Four different problems, four different conversations.
Run These Checks Before the Next Call
Each one is doable without the agency's help, which is the point.
Check whether anything is technically live. Open your key pages and view the source. Is the schema there? Are the rewritten sections published or sitting in a staging queue? A surprising number of stalled engagements are simply unshipped work.
Check whether crawlers can reach you. Confirm your pages return readable content to a plain request. If security tightening happened during the engagement, this could have broken silently in month two and invalidated everything after.
Read the question set out loud to a salesperson. Ask whether customers actually say those things. If they wince, you have found it. Question sets assembled from tools contain industry vocabulary rather than customer vocabulary, and the difference is invisible until somebody who talks to buyers reads the list.
Test five questions yourself, three times over two weeks. If you appear inconsistently, something is working and it is thin. If you never appear, either the target is wrong or the work is not live. If competitors appear consistently, the position is defended and you are contesting rather than establishing.
Check the external listings. If nobody corrected the wrong ones, a substantial part of the work was never done, regardless of what the report says.
Four of those five take under an hour and they narrow the diagnosis considerably before anyone has to defend anything.
Questions Worth Asking in the Meeting
Framed to produce information rather than reassurance.
Which of the four failure modes above do you think this is? A good partner will have a view and will name one. A defensive answer that avoids the question tells you something about how the next six months will go.
What did we deprioritize, and would you make that call again? Every plan excludes things. If nothing was excluded, nothing was chosen, and that is often the underlying problem.
What would you need in order to make this work? Sometimes the honest answer is more budget, and sometimes it is something you failed to supply. Both are useful. An answer of "more time" without a mechanism attached is not.
Show me the actual queries and answers, not the score. You should be able to reproduce three of them yourself. If the reporting is a composite figure with undisclosed weighting, you have no way to verify anything, and independent checking matters more here than in conventional search. Reviewing what any best chatgpt seo analysis tool actually measures is worth doing before you accept a single number as the state of your visibility.
What Realistic Timelines Look Like at Different Scales
Useful context for judging whether month five is genuinely late.
Ken Ganley Mentor CDJR, an Ohio dealership, moved substantially in about two months. Small site, focused local and inventory work, low starting base.
Extension Architecture, a London firm handling extensions and loft conversions, took roughly three and a half months of content clustering, internal linking, and schema expansion against an existing foundation.
EC Council, the global body behind the CEH and CND certifications, needed six months of technical restructuring across a large educational content estate.
Read that spread carefully before concluding you were misled. A large site with a lot of legacy content genuinely takes longer, and the results when they arrive are proportionally larger. A small site showing nothing at month five is a different situation entirely.
When Switching Is the Right Call
Three signals, and they are about behavior rather than results.
They cannot show you the underlying data. If everything is presented as a proprietary score and you cannot reproduce a single query yourself, there is no way to verify anything and no basis for a productive conversation.
The question set came from a tool and they defend it. The mistake is forgivable. Defending it after you have shown them real customer language is not.
Nothing was deployed. If work has been specified for five months and none of it is live, either they cannot manage your development queue or they did not try. Both are grounds to move.
When Switching Would Be a Mistake
Equally important, and less often said.
The blockage was on your side. If your development queue held schema for four months, or nobody supplied the customer question list, or legal rejected content in month three, changing agencies restarts the clock without changing the constraint.
You are month five on a large estate. Switching resets discovery and onboarding, costs six weeks, and puts you further behind than staying.
The early signals are there but small. Inconsistent appearances on lower competition questions is what progress looks like at this stage. It is unsatisfying and it is not failure.
The scope was always too small. If you bought a diagnostic and expected a programme, the mismatch is in the contract rather than the execution. That conversation is about scope, not about partners.
What a Reset Looks Like Without Changing Partners
Often the most efficient path, and it takes one meeting.
Agree a narrowed target. Ten to fifteen questions drawn from actual sales conversations, nothing else, for the next sixty days. Narrow beats broad when you need to prove something is working.
Get everything currently specified deployed before anything new is scoped. Unshipped work is the most common hidden cause and it costs nothing to clear.
Fix the external listings if that has not happened, since it is cheap, high impact, and frequently skipped because it is unglamorous and hard to bill.
Set a checkpoint at sixty days with a defined test you both agree on in advance. If you appear on four of fifteen questions consistently, that is progress worth continuing. If nothing has moved with everything deployed and correctly targeted, the diagnosis is different and you will both know it.
Do the five checks yourself before the next call. Walking into that conversation knowing which of the four failure modes you are in changes it from a complaint into a decision.
Comments
Log in or sign up to join the conversation.