The short answer
An AI marketing agent does one narrow job, usually against your own data, and hands back a work product: a ranked fix list, a content brief, the keywords a competitor ranks for and you don't. That's what an AI marketing agent actually does. It doesn't run your marketing, and it doesn't decide what ships.
The LVL3 Portal has sixteen of them as of November 2026, and every one does a single job and hands back something you can act on. Here's how they split, what a week of runs produces, and where the judgment still sits.
Sixteen agents, four kinds of job
Our agents page groups them by the question you came with. Most read from your own accounts: Search Console, Google Analytics, Google Business Profile, Semrush and keyword data. The rest use a live page fetch, a full site crawl, Chrome field data or an AI assistant.
- Find the next move (4): Keyword Quick Wins, Content Gap Finder, Content Refresh Finder, Keyword Research.
- See where you stand (4): AI Visibility Check, Vertical Benchmark, Competitor Gap Analysis, Backlink Overview.
- Find what is broken (6): SEO Audit, Page SEO Audit, Core Web Vitals, Content Quality, Landing Page CRO Audit, GBP Audit.
- Build the thing (2): SEO Content Engine, Blog Image Generator.
Look at the split. Fourteen of the sixteen find, compare or diagnose. Only two produce anything you'd publish. That's on purpose. The production group is the smallest, and it's where the agents still in development will land.
What an AI marketing agent does in a week
Clients run these themselves, in the same portal we use, so here's a week you could run on one account. Five agents, one a day.
- Monday, Keyword Quick Wins. It reads Search Console and returns the queries where you already rank on page two, sorted by how many people see them.
- Tuesday, Content Refresh Finder. From Search Console and Google Analytics, it finds pages quietly losing traffic they used to earn, each with a brief for what to change.
- Wednesday, Competitor Gap Analysis. From Semrush, it lists the keywords a named competitor ranks for and you don't, filtered by page type.
- Thursday, GBP Audit. It checks every location on your Google Business Profile for name, address, phone, hours and category consistency.
- Friday, SEO Content Engine. Point it at a query (say, one from Monday's list) and it runs keyword research to brief to draft, then hands you an editable document.
By Friday you have five outputs. You don't have a decision yet. Which page-two query deserves the draft, whether the refresh brief is right, whether that competitor is worth chasing at all: that's judgment about your business, and no agent on the list has it.
What an agent doesn't do
- It doesn't set priorities. It narrows the list to what your own data says is worth doing first. Choosing among those is still a person's call.
- It doesn't approve its own work. In our system, automated optimizations are human-reviewed, and on the custom agents we build, a person approves anything that goes to a customer or changes a record.
- It doesn't give a stable answer about AI search from one run. Vertical Benchmark puts a fixed set of your buyers' questions to an AI assistant, and the same question asked twice can come back two ways. The trend is what's worth acting on.
Why the output matters more than the agent
A retainer usually shows you a slide of what somebody ran. The portal shows you the run. These are the same agents we use on your account, in the same place, with the same access to your data, and nothing is queued behind a request.
Runs are saved against your account, so a result from three months ago is still there to compare against. Anything that produces a document gives you a file you can edit. And there is no separate subscription, no per-seat license and no extra login.
How it played out on a real account
The True Food Kitchen engagement (February to August 2026) ran with our portal agents. The work also drew on a 74-item technical, CRO and GEO audit, sixteen months of Search Console data and a 400-row map assigning every target query to a page.
And the 45 location pages that came out of it were still researched one at a time, by people. That's the split I'd point to: agents find and sort, people decide.
Sources
- LVL3 Portal agents page (lvl3.com/portal/agents): the sixteen agents, their groups, data sources and outputs, as of November 9, 2026.
- The LVL3 growth system page (lvl3.com/system) and Custom Agent Development page (lvl3.com/services/custom-agents), on human review, as of November 9, 2026.
- True Food Kitchen case study (lvl3.com/work/restaurant-group), engagement February to August 2026. Published with the client's approval.