The First Ninety Days With An AI SEO Agency

Where It Diverges Sharply Traditional SEO optimises for a ranked list. Generative systems optimise for a synthesised answer, and the sources they pull from are not the same set. Ahrefs studied 15,000 long-tail prompts across four assistants in July 2025 and found that around 80 percent of the pages cited did not rank anywhere for the original query, with only about 12 percent appearing in the top ten. Ranking first does not reserve you a seat.

One thing to establish in week one is where everything lives. The prompt set, the baseline archive, the raw answers and the correction log should sit somewhere you control from the beginning rather than in the agency's systems. Retrieving them later is a negotiation. Having them from the start is an administrative decision nobody objects to at the outset.

A small habit pays off here more than it should. Give each substantial page a short section that states the plain facts in one place: what the thing is, what it costs, how long it takes, who it suits and who it does not. That block is disproportionately likely to be the passage that gets lifted, because it answers several likely questions in a form that survives extraction, and it costs almost nothing to add to a page you were writing anyway.

This frequently produces the first result of the engagement, because access failures are total and fixing them can change answers within days. It should also be short. A fifty page technical audit at this stage is usually padding drawn from a generic template.

Include the Awkward Ones Two categories get left out for uncomfortable reasons and are among the most informative. First, prompts naming your competitors directly, which show whether you appear as an alternative to them.

A prompt set built from internal vocabulary measures how visible you are to people who already talk like you, which is a group that mostly consists of your own staff. It reliably produces flattering results and no useful information.

Consistency Matters More Than Anywhere Else Local identity resolution depends on the business details agreeing across a long tail of directories, many of which nobody has looked at in years. Old addresses, disconnected numbers and previous trading names sit in these places indefinitely.

Most engagements are judged too late, on a final outcome that arrives after the point where anything could have been corrected. The first quarter has its own deliverables, and knowing what they are lets you tell early whether you have hired the right people.

Why Local Is More Exposed The classic local query is a recommendation request with a geographic constraint, and that maps directly onto what a generated answer does well. Somebody asking who to call for a specific job in a specific town receives two or three names rather than a map and a list to work through.

Write Passages That Can Be Lifted Citation happens at passage level, not page level. A model attaches a source to a specific claim, which means the unit of work is a self contained paragraph that remains true and useful when removed from its surroundings.

Who Actually Needs One If your buyers research before they purchase, you are exposed. Software, professional services, healthcare, home services, equipment and anything with a considered purchase all show heavy assistant use at the research stage. If people buy from you on impulse or purely on price at the shelf, this matters far less.

One test of whether a prompt set is any good is to run it and see whether the answers surprise you. A set that returns exactly what you expected is usually measuring your own assumptions, because the questions were written from them. Surprises indicate the prompts reached beyond the company's internal picture of its market, which is the entire purpose.

Weeks One and Two: The Baseline You should receive a prompt set for review, built from your sales notes, support tickets and search queries rather than from your website copy. Read it and check that it sounds like your customers.

That transparency makes it the best available proxy for how retrieval based answering behaves generally. Here is what the citation pattern reveals, and what a brand can actually do about it. llm seo

One more consideration is timing. The cost of entering this channel rises as categories fill up, in the same way that search did between 2005 and 2015. A category with two mediocre comparison articles is cheap to influence today and will not be in three years, once somebody has built the definitive resource and every assistant has settled on quoting it. llm seo

The Mistake Almost Everyone Makes Prompt sets written by marketing teams use marketing language. They contain the category name the company uses internally, the segment labels from the positioning document, and the phrasing from the website.

Month Two: Corrections and the First Rewrites The work should now be concentrated on the recurring sources from the baseline. Expect a list of listings claimed, details corrected and errors submitted, with names and dates attached.