How to train a junior to run GEO clients: a 30-day plan and what not to delegate yet

The AI visibility service has a peculiar growth problem: when it works, it stops fitting in your diary. You run the first accounts yourself because you're the one who understands the topic; by the fifth, you either train someone or you stop selling.

The good news: GEO (optimising a brand's presence in answers from AIs like ChatGPT or Gemini) is more teachable than classic SEO. The discipline is young, the body of knowledge to master is small, and the mechanical part — monitoring, gathering, documenting — is exactly the part that fits a junior profile best. The bad news: the part that isn't mechanical burns accounts if you delegate it too soon.

This is the 30-day plan that junior can follow, and the clear line of what remains yours at the end of the month.

Week 1 (days 1-7): understand the ground

Goal: that by Friday they can explain to their mum what AI visibility is and why an SME should care.

• Days 1-2: the why. Have them read the sector data until they internalise it: frequent ChatGPT use in Spain went from 4% to 28% in two years (Funcas, III AI Survey, 2026); clickless searches are already 69% (May 2025, cited on stucom.com); Gartner forecasts 25% less traditional search before the end of 2026. And the field data: in our study of 9,865 Spanish SMEs, 91% only appear in 1 of the 4 AIs (have them read the whole study, with its sectors and cities: it's the best training material in Spanish right now).
• Days 3-4: the what. GEO vs SEO: what's genuinely new (conversational prompts instead of keywords, 4 engines instead of one, mentions instead of positions) and what's plain old quality SEO. Base vocabulary: that they understand what prompt tracking is (periodically tracking the same questions in the same AIs to measure evolution), why AI answers are volatile and why a one-off screenshot proves nothing.
• Day 5: hands on. Have them run the visibility test on 5 sites: the agency's, two clients and two competitors of those clients. Have them write down what surprised them. That's where you see whether they're hooked or just ticking a box.

Deliverable for the week: a one-pager explaining the results of the 5 tests, in language each business owner would understand. If they write "we optimise citability in generative engines", failed week; back to the vocabulary.

Week 2 (days 8-14): the monitoring machine

Goal: that they master the measurement operation end to end.

• Prompts. That they learn to write the prompts the client's customer would write: natural language, buying intent, local variants. It's the most underrated skill of the service: a bad prompt set invalidates all the measurement that follows. Give them the sets from your current accounts and have them propose 10 new prompts per account; correct them in person, once, with the why behind each correction.
• Tool. That they set up a test workspace from scratch, load prompts, understand the weekly tracking and know how to read the history: when a variation is noise and when it's a trend.
• Basic diagnosis. That they learn the craft's key question: if the brand doesn't appear in an AI, why? A run-through of the sources the AIs draw on (website, structured data, directories, reviews, press) and, very importantly, why each AI says different things about the same brand — because it's the first awkward question a client will ask them.

Deliverable: the prompt set for a real account, reviewed by you, loaded and running.

Week 3 (days 15-21): produce

Goal: that they generate the service's deliverables under your review.

• The first report draft. Give them the structure: the presence figure ("3 of 4 AIs"), the evolution, the findings (new mentions, wrong data detected, competitor moves) and the month's actions. Have them write the report for a quiet account; you correct it in front of them, not by email — the commented correction is half the training. The full audit structure is in the first audit step by step; have them follow it to the letter before they earn the right to deviate.
• Source corrections. That they execute the mechanics: structured data, listing and directory consistency, detection of outdated data. With a checklist and your review.

Deliverable: a complete monthly report for a real account, corrected in session with you.

Week 4 (days 22-30): the account is (almost) theirs

Goal: a supervised handover of 2-3 quiet accounts.

• They run the weekly monitoring and alerts for those accounts.
• They write the reports; you review them before they're sent (this filter doesn't come off until month 3, minimum).
• They attend a reporting meeting with you as a shadow: they present the data part, you carry the conversation. What's shown and in what order is in what to deliver the first month of the service, which serves equally to train the junior and to order the account.

What to delegate now — and what not yet

Delegable from day 30:

• The full monitoring: tracking, alerts, history, documented screenshots.
• The first draft of the monthly report (with your review before sending).
• Prompt research and its maintenance.
• The mechanical source corrections, with a checklist.
• The descriptive part of the meeting: "this is the month's data".

Not delegable yet (and not for quite a few months):

• The hard conversation. "We've dropped in ChatGPT and we're not sure why" needs someone who can hold the uncertainty without over-promising or sinking. A junior, faced with a tense client, promises. And in a channel where nobody controls the AI's answers, promising is mortgaging yourself.
• The proposal and the price. What to include, how much to charge, which KPIs to commit to and which only to report: that's commercial judgement, slow-cooked. The junior can prepare the proposal's data; you sign the proposal.
• The strategic reading. Telling "this is weekly noise" from "this is a trend that demands action" is what the client really pays for. It's learned by seeing many accounts over many months; it's not in any checklist.
• Promise control. Review every text that goes out to the client looking for one thing only: implicit guarantees. "We'll get you to appear" can't be written, not even with the best intentions.

Frequently asked questions

What junior profile works best for this?

Better writing than technical skill. The service's day-to-day is writing prompts the way people speak, interpreting text answers and writing clear reports. A humanities profile with curiosity about AI usually performs sooner than a technical one who writes for machines.

How many accounts can a trained junior handle?

With automated monitoring, the limit is the interpretation and reporting work: between 6 and 10 accounts from the third month, depending on each service's intensity. Without a tool, with manual monitoring, not even half — and burned out within a quarter.

What if the junior gets trained and leaves?

A classic risk with a classic answer: document the process (this plan is the start), spread the knowledge as soon as there's a second person and remember that the alternative — training nobody — leaves you as the bottleneck forever. The only truly serious thing is for the judgement to leave without ever having been written down.

When can they start running the reporting meeting on their own?

Once they've done two meetings as a shadow and a third leading the voice with you present in silence. If in that third one they promise nothing impossible and can say "I'll have to check that" without getting nervous, they're ready for the quiet accounts. The hard ones stay yours.

---

For the plan to work, the junior needs the mechanical part solved: with Surfeo for agencies, each client has their workspace with the 4 AIs monitored every week, and your junior spends their time learning what actually matters: interpreting and telling the story. Their first exercise can start today: the free visibility test on any client's site.