What to expect from an AI SEO agency in the first 90 days: an audit with first fixes live in week one, technical cleanup and rebuilt priority pages in month one, and content shipping steadily through months two and three, with early movement showing up in impressions before it shows up in clicks. That's the real shape of a good engagement. Anything dramatically faster is usually a lie, and anything dramatically slower is usually onboarding theater with a retainer attached.
We'll walk the timeline using our own three-phase process, the Reforge Method (Assay, Reforge, Temper), but the shape generalizes. Whoever you hire, hold them to something like this.
Week one: demand work, not onboarding
By the end of week one you should have two things: a real audit and at least one fix live on your site. Access setup happens here too (analytics, Search Console, the CMS), but access is a Tuesday task, not a milestone. Any agency that treats week one as pure paperwork is billing you for its own admin.
The audit (our Assay phase) covers three layers. Technical: crawlability, indexing, structured data, anything silently blocking retrieval. Search data: what you already rank for, which pages earn impressions, where the fastest wins hide. And AI answers: what ChatGPT, Perplexity, and Google's AI Overviews currently say when asked about your category and your brand. That last layer is the baseline everything at day 90 gets measured against, so if the agency never records what AI assistants say on day one, they can never prove they changed it.
And there is always something broken enough to fix in week one. A blocked page, a chain of redirects, a money page with a title nobody would click. If week one ends with kickoff decks and zero shipped changes, you've just learned something about the next 89 days.
Month one: fix the foundation, reforge the money pages
Month one is triage plus rebuilding. The audit becomes a ranked fix list, and technical items go first because everything else sits on top of them. Indexing problems, crawl waste, missing schema, broken internal links. Unglamorous and decisive.
In parallel, your priority pages get reforged: rebuilt answer-first, with a direct answer in the opening lines, question-shaped headings, and structured data that makes the page legible to machines. Priority means the pages closest to revenue, not the blog. One services page that answers plainly beats ten new posts nobody asked for.
What you shouldn't see in month one: big traffic jumps. What you should see: a visible trail of shipped changes, each one explained in plain language.
Months two and three: ship content, start earning authority
Now the cadence sets in. New pages and articles ship on a steady schedule (at our published tiers, that's 4 or 8 per month), targeting questions your customers ask. Comparison pages, plain pricing answers, specifics an assistant could quote. Volume for its own sake proves nothing; a page only counts if it answers something. You should be able to open any shipped page and say what question it exists to win.
Authority work starts here too, our Temper phase: earning placements and mentions on sites that matter, so citations stick to a brand machines recognize. Placements land slowly because pitching real editors takes real time. A pipeline with named targets is normal at day 90. A pile of mystery links is not, and the difference matters because bad links are a liability you'll pay to remove later.
By day 90 the agency should also rerun the week-one AI answer tests. Same questions, same engines, results side by side. We cover what moves those answers in how to get cited by ChatGPT.
When do results show up?
Impressions move before clicks. Rebuilt pages typically start appearing for more queries at low positions first: you surface on page two or three for terms you never ranked for, collect impressions, and only later climb into positions that earn clicks. So a normal 90-day chart shows impressions climbing, rankings wobbling upward, and clicks just beginning to follow. The wobble is worth naming because it scares people: a page that jumps to position 8, drops to 15, and settles at 6 over three weeks is behaving normally. Search engines test new pages against user behavior before committing to a position.
Content-driven rankings and AI citations build through months 3-6 and compound after that. We say this before taking anyone's money, because the alternative (promising page one in 30 days) requires either lying or spam. Learn to read the early signals yourself in how to measure AI search traffic, so you're not dependent on anyone's interpretation of your own data. Including your agency's.
What should reporting look like from day one?
Three artifacts, none optional:
- A weekly email: what shipped, what moved, what's next. Short enough that you'll read it. An agency that can't fill it weekly isn't doing weekly work.
- A live dashboard: rankings, traffic, AI citations, every page touched. Built on your accounts (your Search Console, your GA4) so the data outlives the engagement and nobody grades their own homework.
- A monthly review: what compounded, what didn't, what gets reprioritized. Real reporting includes the second category.
One test covers all three: if you canceled today, would you know exactly what you paid for? With good reporting the answer is yes at any moment of the engagement. If the answer is "sort of," the reporting is marketing.
Is it a red flag, or just normal slowness?
Both exist at day 90, and confusing them ruins good engagements while prolonging bad ones.
| Normal at 90 days | Red flag at 90 days |
|---|---|
| Rankings wobble while impressions climb | Nobody can list what shipped, with dates |
| Clicks lag weeks behind impressions | Dashboard access is still "being set up" |
| Few AI citations on competitive queries yet | Baseline AI answer tests were never run |
| Authority placements still in the pitch pipeline | Links appear from sites no one can explain |
| Leads not yet attributable to organic | Guarantees come out when you ask for specifics |
The left column is compounding on schedule. The right column is the catalog we cover in the problems with hiring an AI SEO agency, surfacing on schedule. The rule that separates them: patience is owed to transparent slowness, never to opaque activity.
What should you do at day 90?
Hold a review with three questions on the table. What shipped, versus what the package you're paying for promised? What do the week-one and day-90 AI answer tests show side by side? And what's the plan for months 4-6, when the compounding is supposed to arrive? Good agencies bring this review to you before you ask for it.
And if you're still evaluating agencies, run the week-one test yourself before you sign anything. Write down the ten questions your customers ask, run them fresh in ChatGPT and Perplexity, and record who gets named. It costs an afternoon, any agency worth hiring will be glad you have the baseline, and at day 90 it will tell you whether you hired the right one.
We've packaged this whole discipline as a free First 90 Days Tracker: the 13-week checklist with an evidence column, the week-one AI baseline template, and the day-90 review agenda. It unlocks with an email, it works with any agency including us, and holding vendors to timelines is precisely the kind of tool we think buyers should have.