Marketing Brain.Get Found by AI.
Then measure it. A three-phase engagement: we audit where you appear when buyers ask ChatGPT, Claude and Perplexity instead of Google, rebuild the site so both people and models can get a straight answer out of it, and hand you a private dashboard — wired to your analytics, search, session, identification and CRM data — that you can query in plain language from inside Claude.
Two things broke at the same time.
The first is discovery. A growing share of buyers now open ChatGPT, Claude or Perplexity instead of a search box, and those systems return one synthesized answer built from a handful of cited sources. Ranking on page one is no longer the same thing as being in the answer. You can hold position three for a term and be entirely absent from the paragraph a model writes about it.
The second is measurement, and it is the quieter problem. Most marketing stacks cannot see any of this. Answer engines send little attributable referral traffic, analytics properties routinely have no conversions configured at all, and the tools that identify who visited keep no history you can go back and read. The exact moment discovery changed shape is the moment the dashboard went dark.
This package fixes both, in order, because the sequence matters. There is no point optimizing for citations you cannot measure, and no point building a dashboard on top of a site that has nothing extractable to say.
Three phases.
Each one is a standalone deliverable with its own finish line. Buy the first and stop, or run all three.
GEO & AEO Audit
We measure where you actually appear when a buyer asks an AI instead of a search engine — and where your competitors appear instead of you.
- —Prompt coverage across ChatGPT, Claude, Perplexity and Google AI Overviews, scored against a named competitor set
- —Citation gap analysis — which sources the models quote in your category, and why none of them are you
- —Crawler access audit: robots rules, rendering, and whether AI crawlers can read the page a human sees
- —Structured data and answer-shape review — schema coverage, question-shaped headings, extractable answers
- —A measurement baseline, so phase three has a before to compare against
A prioritized fix list with the baseline attached. It stands alone — you can hand it to your own team and stop here.
Website Refresh
The audit finds the gap. This closes it. We rebuild the pages that matter so both a person and a model can get a straight answer out of them.
- —Page-level rewrites into answer shape: the question asked plainly, the answer in the first paragraph, evidence under it
- —Full structured data pass — Organization, FAQPage, Article, Breadcrumb — validated, not just added
- —Performance work against Core Web Vitals, because a page a crawler times out on is a page that does not exist
- —Conversion path repair: the route from an AI-referred visitor to a booked call, instrumented end to end
- —Shipped on Next.js and Vercel, in your repo, under your account
A site your buyers can find and your analytics can see. No lock-in — you own the code.
The Marketing Brain
A private dashboard that reads your live marketing data, plus an MCP server so you can ask it questions from Claude or ChatGPT instead of reading charts.
- —Nine screens: overview, acquisition, audience, people, explore, press, health, audit and connect
- —Five sources wired in — analytics, Search Console, session recording, visitor identification and your CRM
- —A warehouse for the sources that forget: identification and session tools keep no history, so we keep it for them
- —A remote MCP server with 19 tools, so the same numbers answer in the dashboard, over the API and inside a chat
- —Per-person API keys with scopes and an audit trail — no shared secrets, revocation is immediate
- —Scheduled jobs with a heartbeat that alerts on silence, because a dead feed looks exactly like a quiet week
Deployed in your infrastructure, on your accounts, with a per-person handover doc. Then a monthly retainer keeps it accurate and keeps working the visibility.
What the Marketing Brain actually is.
Six surfaces over one set of definitions. The dashboard, the chat, the API and the alarm all answer the same question the same way.
The dashboard
Nine screens your team opens in a browser. Every panel names its source and its freshness. The health screen exists specifically to tell "no data" apart from "not connected" — the distinction that earns trust in the first week.
The MCP server
Nineteen tools over a stateless remote MCP server. Ask Claude or ChatGPT which companies from paid search came back twice this month, and it queries the same definitions the dashboard uses. One registration serves both transports, so the two cannot drift.
The panel in the chat
On hosts that support the MCP UI extension, a tool can return the dashboard itself — rendered in a sandboxed frame inside the conversation, not a link out to it. The underlying tools still stand alone everywhere else.
The API
The same runners over plain HTTP with per-person bearer tokens, scoped at mint time. Every request is written to an audit trail. When someone leaves, you revoke one key.
The alarm
An hourly heartbeat watches the push-fed tables and alerts to Slack when one goes quiet. It fires on state change, not on a schedule, so it stays an alarm rather than noise you learn to ignore.
The audit screen
Every scheduled job writes a row even when it does nothing, so silence becomes visible. This is the screen that catches the failure where a feed dies and the charts keep looking plausible.
Rules the dashboard cannot break.
Every marketing dashboard we have been asked to replace had the same defect: it was confidently wrong somewhere, and nobody could tell where. These rules are written into the codebase and enforced by an automated checker that has to pass with zero failures before anything reaches a client. Each exists because the opposite behavior shipped once and produced a number someone nearly acted on.
No mock data. Ever.
If there is no live query behind a number, the number is not there. Not a placeholder, not a sample, not a reasonable-looking estimate that survives into the board deck.
Null means not measured. It never means zero.
An unconfigured source renders as unconfigured and names the credential it is waiting on. The failure we refuse is a screen reporting a confident zero for a feed nobody ever connected.
Two sources are never averaged.
Analytics and session recording will disagree about the same week. We show both and say which is the source of record, rather than splitting the difference into a figure that is true nowhere.
A dead source costs one panel, never the page.
Every adapter wraps its own failure. A revoked token takes down the tile that needed it and nothing else, and the health screen says so in words.
Time is bucketed in the source timezone, and the screen says which.
Silent timezone drift is how a Monday number becomes a Sunday number. Every date range declares the zone it was cut in.
A coverage gap is drawn on the chart.
When a feed was down the trend line shows a gap, not a dip. One client lost fifteen days of identification data and nobody noticed for a month. That cannot happen twice.
What we find, almost every time.
Four patterns so consistent we now raise them on the first call, because each one changes the shape of the engagement.
Your conversions are probably undefined, not zero
Most analytics properties we open have no key events configured. The site fires the events; none is marked as a conversion. Every channel is therefore credited with nothing, and the report reads like your marketing does not work. It is a config change measured in hours, and it is usually the highest-value fix on the engagement.
Identification and session tools cannot backfill
Every day before we have access is a day that will not exist later. There is no history to recover once it is gone. We say this on the access call because it reliably moves a decision from next quarter to that afternoon.
Enrichment is wrong on exactly the rows that matter
Visitor identification gets the company wrong roughly a fifth of the time in our own measurement. If a lead score derives from company data, a wrong company moves that person up the call list. We treat identification as signal, never as ledger, and the interface says so.
Most of your search traffic already knew your name
On one recent build, 99.7% of organic search was branded — four non-branded clicks in twenty-eight days. That is demand capture, not discovery, and it is the clearest argument we know for why answer-engine visibility is now a separate job from SEO.
What it costs.
Published, because you should be able to tell whether this is in your range without booking a call to find out.
GEO & AEO Audit
Fixed scope, two weeks, prioritized fix list and baseline. Stands alone.
Website Refresh
Priced against the audit findings and page count, so you see the number once we both know the work.
Marketing Brain
One week to build, then a retainer covering accuracy, new sources and ongoing visibility work.
Phases are sequential but not bundled — the audit is the entry point, and most clients decide on the rest once they have seen it. Training your team to run the Brain themselves is included in the handover, not billed as a separate engagement.
Answered before the call.
Start with the audit.
Two weeks, fixed price, and you will know exactly where you stand when a buyer asks an AI about your category. Everything after that is your call.