The monthly content plan for a GEO client: what to publish so the AI cites you
You've signed the GEO client —GEO, Generative Engine Optimization: working on their presence in the answers of AI search engines like ChatGPT, Gemini or Perplexity— and the baseline is done. Now comes the question that decides whether the service works: what do you publish each month so the AIs start citing them?
Because AIs don't cite pretty websites or long texts: they cite pieces from which they can extract a specific answer. In the study we ran on 9,865 Spanish SMEs across 30 sectors and 10 cities, 91% appeared in only 1 of the 4 AIs (full study), and the pattern repeats: most have published nothing an AI can use to answer. The GEO content plan exists to manufacture exactly that, month by month.
The four types of piece AIs cite
1. Honest comparisons. "Best X in Y", "X or Z, which to choose". People ask the AI in comparative form, and the AI builds its answer from published comparisons. The uncomfortable condition: it has to include real competitors and treat them fairly, because a "comparison" where your client wins at everything isn't citable, it's a brochure. One comparison a month that genuinely helps a decision is worth more than four pieces of self-promotion.
2. FAQs with direct answers. The literal question as a heading and the answer resolved in the first two sentences, with no introductory preamble. It's the easiest format for a machine to extract and the one that covers the long tail of real doubts: "how long does...?", "do I need an appointment...?", "what does it include...?". The raw material is free in the client's day-to-day: what people ask them over the phone and on WhatsApp.
3. Pieces with concrete, verifiable data. Prices from, timeframes, the business's own figures. AIs cite numbers before adjectives: "from €49 and delivery in 72 hours" is citable; "competitive prices and fast delivery" is invisible. If the client has their own data —cases, statistics from their activity, measured results— that's gold: no one else can publish them and the AI can only cite them.
4. Local pages. For a business with a geographic scope: pages by service and area with address, opening hours, coverage and local particulars. A huge share of questions to AIs carries local intent ("near me", "in Granada"), and the AI answers with whoever has that information published in an extractable way.
Notice what's not on the list: the generic "5 trends in [sector]" post that contains not a single piece of its own data. That content adds nothing the AI doesn't already have from a thousand sources, and publishing it is the most expensive way of not advancing.
The realistic cadence: where 6-16 articles a month come from
The cadence isn't set by inspiration but by the audit: each week of monitoring tells you which questions still have no published answer and where the competitor is being cited. That's the backlog.
On volume, real numbers: in Surfeo, each client generates 6 article drafts a month on the Starter tier and 16 on Growth, written from the gaps their own audit detects. A typical month with 6 pieces looks like this:
- 1 comparison on the client's main purchase decision.
- 2 FAQ pages attacking the groups of uncovered questions with the most presence in the audit.
- 2 data pieces (prices and timeframes for a service, a case with figures).
- 1 local page for the next service+area on the map.
With 16 pieces the recipe doesn't change, the pace does: you cover the question map in three months instead of eight. For a client in a competitive sector, that difference shows in when you start being able to show results.
The role of the human: reviewing isn't optional
The automated draft solves the blank page; it doesn't solve the truth. Before publishing, someone on your team who knows the client spends 20-30 minutes per piece on three things:
- Accuracy: prices, timeframes, services and claims checked with the client. This is the critical part: if you publish a wrong fact, you're teaching that error to the AIs that will then repeat it.
- What only the business knows: the anecdote, the nuance, the operational detail no generator can invent. It's what turns a correct piece into a citable one.
- Publishing criterion: if the piece doesn't add a fact or an answer that wasn't already published on the client's website, it doesn't get published. Volume without novelty is noise.
Where this review falls within the agency's week —and what else is worth automating or not— is covered in what to automate and what not in a GEO service.
How you'll know if the plan is working
Not by the blog's traffic: by coverage. Month by month, what percentage of the client's key questions already have a published answer, and in how many of them the AIs mention them. These are different metrics from classic SEO, and it's worth knowing which you can commit to in the proposal and which only report — the list is here. The reward of doing it well has a number behind it: traffic arriving from AI answers converts at 14.2% against 2.8% for classic organic, according to sector data collected by roymo.es. Fewer visits, better visits.
Frequently asked questions
How many articles a month does a client really need?
It depends on the size of their question map and the competition. A local business with 40 key questions progresses well with 6 pieces a month; one in a competitive sector or with many service lines justifies 16. Beyond that, the bottleneck stops being generation and becomes human review: if you can't review, don't raise the volume.
Doesn't Google penalise AI-generated content?
Google has repeatedly said it assesses the quality and usefulness of content, not the tool it was made with. The real risk isn't a penalty: it's publishing filler with no data, which neither Google rewards nor AIs cite. That's why the human editorial criterion in the previous point isn't an adornment of the service, it is the service.
What do I do if the client has no data of their own to publish?
They have it; they just haven't written it down. Real prices and timeframes, the 20 most repeated questions on the phone, what their typical client asks for, figures from their activity. The first content meeting is an interview to extract that. If there really is nothing, start with FAQs and local pages, which are built from basic operational information.
When do the first results show?
AIs have no fixed calendar and promising dates is selling yourself short. The honest answer: changes in low-competition questions can show in weeks; in competitive questions, months. That's why the first month delivers baseline and plan, not mentions — what to deliver exactly in the first month is here.
Want to see a client's starting point before building their plan? Run the free AI visibility test: in minutes you'll know which questions they appear in, which they don't, and where their first month of content should start.
