AI capability your team keeps

AI capability

Most marketing teams are using AI to write faster. That’s the smallest thing it does. The bigger one is that the research, the measurement, and the quality control your team never quite has time for can run as systems instead of as tasks somebody has to find a spare afternoon for. That’s how a four-person team covers ground a larger one covers.

I build those systems inside your operation, and I own them while they run. You keep them.

What this catches that a person working alone would miss

The point isn’t speed. It’s that a system reads every row, every time, and a person reading a report reads the summary. Here’s what that difference has actually turned up. Clients are described by industry, because much of this runs through agency partners and those relationships aren’t mine to name.

• 113 rankings that weren’t rankings. A sports advertising company’s rank tracker showed 113 keywords in position 1. Only 3 were true first-place organic results. The other 110 were mentions inside Google’s AI Overviews. I rebuilt the content strategy around earning citations rather than chasing positions, put measurement in place to show whether those citations were helping or hurting the business, and ran fresh keyword research against the business concepts their competitors hadn’t touched.
• 85 forgotten pages earning real traffic. Comparing Search Console against a full site crawl surfaced 85 legacy URLs that nothing on the site linked to any more. They were still bringing in roughly a quarter of the brand’s organic clicks. I consolidated some, retired others, and put the rest on a refresh calendar to rebuild both their rankings and their presence in AI answers. Organic visibility is up 20%.
• An audience definition wrong by two orders of magnitude. A “repeat purchasers” segment was filtering on order value instead of order count, so it treated about 90% of buyers as repeat customers. Rebuilt correctly, it found 3. That single fix moved paid spend on Google and Facebook onto a different strategy, and the email program was refocused around positioning that matched who the customers actually were.
• A lead count that meant nothing. For a manufacturer, monthly lead reporting was mixing job inquiries and vendor pitches in with real sales leads. I rebuilt the measurement model in Google Tag Manager and GA4, separated real leads from every other inquiry type, and added a Google Data Studio reporting layer so the client reads clean numbers instead of interpreting them.
• 413 keywords sorted the wrong way. For a precision manufacturer, the tracked keyword set was grouped by SEO technique rather than by what the company can actually sell. I rebuilt it as a four-tier system, with Commercial Priority at the top and high, medium, and low opportunity beneath it, merging variants and pulling in striking-distance terms, then added a scoring mechanism that sets the priority order for each quarter’s content plan.

Notice what those five have in common. The finding is the cheap part. What it’s worth is the change that follows, and that’s the part that needs somebody senior still holding the account when the answer comes back.

What I build inside your stack

Anything I do twice becomes a system. These are the ones that carry from one engagement to the next, set up against your data and your definitions rather than generic ones.

• A keyword and opportunity library. Rank tracking, Search Console, competitor data, and a full site crawl pulled into one workbook, scored and grouped so a strategist can act on it. It carries a content opportunity score and a decay score, so you know what to write and what to refresh.
• A content brief system. Four documents per piece: a strategy note, a writer packet, an SEO pass, and a CMS load sheet, so each person gets only what they need. A coverage check runs before anything is commissioned, which stops you paying for a post that competes with a page you already have.
• A monthly opportunity run. Fifteen or more content ideas screened against four tests: is it on-service, can it win against what’s ranking now, does it fit a cluster with buying intent, and does the evidence hold up.
• Measurement and reporting pipelines. GA4 and Tag Manager audited and rebuilt, then proven on the live site event by event rather than trusted from a settings screen.
• A site evaluation framework. Content, data, technical, and authority tracks, with every criterion tagged for classic search, AI answer engines, or both.
• A five-phase production system. Content moves from research through to publication in five defined phases, with CMS delivery criteria and gated checks at each one, so nothing ships without the documentation and the process behind it holding up.
• Scheduled research agents. They scan their sources, confirm every result is still live before reporting it, and log what they found.

What leverage actually looks like on your team

Capability on its own is a delivery. Leverage is what happens after. Your team stops spending its week on the work a system should be doing, and starts spending it on the work only they can do: talking to customers, making the call on positioning, getting things shipped.

The reason that keeps working is ownership. Channels change, field definitions drift, a tracker gets renamed, an export changes shape, and a system nobody is watching goes quietly wrong. Somebody senior has to be reading the output and adjusting the rules behind it. That’s the part I keep doing every month, and it’s why this is an ongoing engagement rather than a build you buy once.

How I keep AI honest

AI is fast and it sounds sure of itself, which is exactly why it needs checking. These are the rules I hold my own work to.

• Every number traces to a source. If I can’t point to the export or report a figure came from, it doesn’t go in your deliverable.
• Big builds pass automated checks first. Keyword workbooks aren’t complete until a verification suite passes. One client library ran 52 checks before delivery.
• Inputs get checked before the work starts. A filtered or truncated export gets flagged at the beginning, not discovered at the end.
• Live data beats old notes. I re-read the current file or account instead of trusting what it said last month, and I log corrections so the same mistake can’t come back.
• Limits are stated plainly. Every deliverable says what the data can’t tell us yet and what I’ll be watching.

A real example. In one build, the checks caught a traffic figure that had come from a crawl tool’s partial data instead of the full 12-month Search Console export. I corrected it in every file where it appeared.

AI and your data

Most teams have been asked about this by their own legal or security people, so here’s my position rather than a request for yours. I work in Claude, Anthropic’s AI model. Your data isn’t used to train it. What I extract for analysis stays in the workspace for your engagement, and I review every output before it reaches you. If your company has its own rules about AI and client data, tell me at the start and I’ll work inside them.

Where the work lives. I build in my own Claude workspace, because that’s where my tooling is and it keeps your account clean while things are still moving. Once a system is working and you have Claude Team or Enterprise, I rebuild it inside your workspace, so it runs on your account, under your admin controls, with your people able to open it. If you don’t have that yet, I document it so it can be stood up whenever you do. Either way you finish the engagement owning the systems rather than renting access to mine.

What stays with you

Everything I build for you is documented while it’s being built, not written up afterward. The workbooks, the briefs, the tracking setup, the reporting, and the process notes that explain how each one runs are all yours. If we stop working together, you keep them and your team can run them.

What I keep is the method, which I bring to the next client the same way I brought it to you. What I don’t keep is your data, your content, or anything built against your business.

None of that is new, and it isn’t really about AI. It’s how I’ve worked since long before these tools existed. Someone who watched me do it in-house put it this way:

Questions I get asked

Isn’t this just using ChatGPT to write blog posts?

No. Drafting is the least of it, and I don’t hand you AI-written copy. The work is in the systems that decide what’s worth writing, prove whether the tracking is telling the truth, and catch the things a summary hides. The 113-rankings example above took a system reading every row.

Does my team need to learn anything new?

Only to read the output, which is built to be readable. The systems run on my side and report into yours. If someone on your team wants to take one over, that’s a good outcome and I’ll teach them.

What happens to all this if we stop working together?

You keep it. It’s documented as it’s built, so there’s no handover project at the end and nothing that only works while I’m holding it.

How do I know the numbers are right?

Every figure traces to a named source, large builds pass a verification suite before delivery, and each deliverable states what the data can’t tell us yet. When a check catches something, I correct it everywhere it appeared and log it.

Is this a separate service or part of an engagement?

Part of one. It isn’t sold on its own, because a system without an owner goes stale inside two quarters. It comes with fractional leadership or a retained specialty engagement.

Want this running on your marketing?

Tell me what your team is spending its week on and where you suspect the numbers aren’t telling you the truth. I’ll tell you plainly whether I’m the right fit.

Or see how this fits into fractional marketing leadership.