Building and Staffing a GEO Practice
A GEO practice is four things working together: a small set of people with the right skills, measurement you trust, an experimental loop that runs on a cadence, and a direct line into product planning. Tools are the least important of the four.
Chapter 9 assigned ownership. This chapter is about making that ownership operate week to week.
Buy the commodity, own the product
Start with what not to build. AI visibility tracking, prompt sampling across engines, and citation monitoring are becoming commodities, with many vendors offering similar capabilities. Buy that layer or use a capable free tool. Chapter 12 helps you choose.
What you cannot buy is the material only your company has: the prompt set drawn from your own deals, the truth sheet, the product fact model, the experiment log and the domain judgment to interpret it all. Those are the practice. A team that owns them can switch tools in a week. A team that outsourced them has nothing left when the contract ends.
The roles
A practice needs these capabilities. In a small company one person covers several; in a large one each may be a team.
| Role | Owns | Core skills |
|---|---|---|
| GEO lead | Prompt set, truth sheet, priorities, reporting | Product marketing, domain knowledge, cross-team influence |
| Measurement analyst | Sampling design, dashboards, experiment analysis | Statistics, analytics, skepticism about single numbers |
| Product-layer liaison | Naming, fact model, exposure backlog in product planning | Product management, enough engineering to scope work |
| Docs and developer relations | Documentation coverage, accuracy, agent readability | Technical writing, API fluency |
| Content strategist | Content program aimed at citation gaps | Research-led writing, vertical expertise |
| Technical SEO | Crawlability, rendering, crawler policy, markup hygiene | Technical SEO, web engineering |
| Comms | Earned coverage, analyst and review programs | PR, analyst relations |
The first three are the new ones. The last four usually exist already and need a new brief, not a new hire.
What to hire for
When you hire or assign someone to lead GEO, weigh three things above tactical experience.
Domain depth. Someone who understands your buyers' vocabulary, standards and authority sources will design better prompt sets and spot wrong descriptions faster than a generalist with a longer GEO resume. In security and identity, this is decisive.
Experimental thinking. Ask candidates to design a test for a single hypothesis, such as whether adding functional terms to feature names raises mention rate. Listen for controls, repeated sampling and time windows. Be wary of anyone who promises a ranking or a citation from a tactic.
Product fluency. The lead will spend much of their time persuading product teams to change names, docs and data. They need to speak in roadmaps and tradeoffs, not marketing metrics.
Measurement infrastructure
Measurement comes in three tiers. Most teams only have the first.
Tier 1: answer monitoring. A fixed prompt set, sampled repeatedly across the engines your buyers use, tracking the metrics from Chapter 3: mention rate, citation rate, first-pick rate, description accuracy and source mix. This is what visibility tools provide. Add your own accuracy scoring against the truth sheet, which most tools do not do well.
Tier 2: traffic and behavior. AI referral traffic segmented as its own channel, and AI agent requests segmented in server logs. Standard analytics setups often lump AI referrals into direct or generic referral traffic. A GoodFirms survey in May 2026 found that only 14% of marketers track AI search citations at all (GoodFirms). Fixing the analytics side is usually a few days of work.
Tier 3: pipeline. Connect AI-influenced visits and self-reported attribution to opportunities. Add "AI assistant (ChatGPT, Claude, Perplexity, Gemini, Copilot)" as an option in your how-did-you-hear field, and ask about it in discovery calls. Much AI influence never produces a click, so self-reported data fills a gap that analytics cannot.
The inversion in the Introduction only becomes visible with all three tiers. Tier 1 shows citations rising. Tier 2 shows traffic falling. Tier 3 shows whether the buyers who do arrive are better.
The operating cadence
A practice runs on rhythm, not campaigns.
| Cadence | Activity | Output |
|---|---|---|
| Weekly | Review answer monitoring; triage wrong or missing descriptions | Tickets for product, docs or content |
| Every two weeks | Experiment check-ins; start or stop tests | Updated experiment log |
| Monthly | Cross-layer review with all owners; prioritize the product-layer backlog | Priorities for next month, escalations to product planning |
| Quarterly | Refresh the prompt set from recent deals; verify the fact model; consistency audit of external surfaces; rerun the description test | Baseline update and quarterly report |
| Every release | Definition-of-done check: naming pairing, docs at launch, fact model updated | No new product-layer debt |
The release-level check matters most over time. Without it, every new feature adds to the backlog you are trying to work down.
Getting product-layer work done
The practice will generate product-layer tickets: rename lines, missing docs pages, catalog fields, exposure work. They must live in the product team's tracker, not a marketing spreadsheet, or they will not ship.
Write them like product tickets. State the problem as evidence ("Engines describe SSO as an enterprise add-on in 7 of 10 runs; it is included in the business tier"), the change, the surfaces affected, and how you will measure the result. Small, evidence-backed tickets get prioritized. Requests to "improve AI visibility" do not.
The first 90 days
A new practice can show real progress in a quarter.
- Days 1 to 30: baseline. Write and sign off the truth sheet. Build a prompt set of 50 to 100 prompts from recent deals, segmented by vertical and buying stage. Run the description test across the major engines with repeated sampling. Fix AI referral segmentation in analytics. Score the five audit layers.
- Days 31 to 60: first experiments. Run two or three controlled tests on the cheapest high-impact fixes, usually feature naming pairings and two or three missing docs pages. Start the product fact model with identity, capabilities, standards and integrations.
- Days 61 to 90: make it routine. Publish the canonical descriptions and update owned surfaces. Start the weekly and monthly cadence. Add the definition-of-done check to the release process. Report the baseline, the first results and the next quarter's priorities to leadership.
Scaled to company size
Startup. One person spends part of their week on GEO, usually in marketing, with the founder or head of product as the product-layer owner. Buy a tracking tool, keep the prompt set small and sharp, and focus almost entirely on naming, docs and the truth sheet. The founder's involvement is the advantage: decisions that take months in a large company take a conversation.
Mid-sized company. A dedicated GEO lead in product marketing, part of an analyst's time, and named liaisons in product and docs. The monthly cross-layer review is the heartbeat.
Enterprise. A program office with a lead, an analytics function, liaisons in each product line and governance over the shared prompt set and fact model. The risk is many teams running separate GEO efforts with different numbers. Insist on one measurement system.
What to report to leadership
Leaders need a few numbers that connect to the business, not a dashboard of every engine.
- Description accuracy on the truth sheet, across engines, over time
- Mention rate on pipeline prompts, the prompt clusters drawn from real deals
- AI-influenced pipeline, from analytics and self-reported attribution
- Product-layer debt, the open backlog of naming, docs, data and exposure fixes
Label trends as trends and keep confidence intervals in mind. A practice that reports noise as progress loses credibility the first time the noise reverses.
Common failure patterns
- Tool first, practice later. Buying a dashboard before deciding what to measure and who will act on it.
- Content treadmill. Responding to every visibility gap with another article.
- Orphaned tickets. Product-layer fixes that live in marketing documents and never reach product planning.
- Prompt sets that never change. Buyers' language moves; a prompt set from last year measures last year's market.
The question for this chapter: if your GEO lead left tomorrow, which of the practice's assets would leave with them?
Key takeaways
- Buy the commodity (tracking and sampling) and own the product: prompt set, truth sheet, fact model and experiment log.
- Hire for domain depth, experimental thinking and product fluency over tactical GEO experience.
- Measure in three tiers: answer monitoring, traffic and behavior, and pipeline.
- Run on a cadence, put product-layer fixes in the product team's tracker, and add an AI legibility check to every release.