AI SEO: A Practical 30-Day Operating Routine for Lean Teams
A step-by-step guide to using AI to scale SEO work and to get your brand cited in AI answers, built as a repeatable monthly cadence rather than a one-off project.
AI SEO Is Two Different Jobs - Naming Them First Saves Months
If you run a small marketing team, "AI SEO" probably arrived as a vague instruction: do something with AI for our search visibility. The fastest way to waste three months is to treat that as one project. It is two jobs that share tools but not goals.
Job one is using AI to do SEO work faster: keyword clustering, brief drafting, internal link suggestions, metadata variants, performance triage. Job two is optimising so that AI answer surfaces - Google AI Overviews, ChatGPT, Perplexity - cite your brand when someone asks a category question. Job one is about output volume and speed. Job two is about presence in someone else's generated answer.
Naming them matters because they have different success metrics and failure modes. A team that blurs them generates 400 pages nobody asked for (job one without a gate) while never checking whether AI Overviews cite them at all (job two never started). Keep two scoreboards: pages published and ranked, and citation share won in AI answers.
Both jobs are legitimate. Google's guidance states automation is not against its rules and the standard is helpful, reliable, people-first content. The line is about purpose and value, not whether a human or machine typed the first draft.
The rest of this guide is a 30-day operating routine for a 2-4 person team: prerequisites, six sequenced steps, a written publish gate, five mistakes that stall small teams, and a monthly checklist. It assumes you have a website with some history and nobody is going to build you a data pipeline.
What a 2-4 Person Team Should Have in Place Before Starting
You do not need new software. You need four things agreed and written down, because retrofitting them mid-cycle causes rework.
| Prerequisite | What "done" looks like | Who owns it |
|---|---|---|
| Written quality and brand rules | One page: tone, claims you can and cannot make, banned phrases, required source types | Marketing lead |
| A place to publish | A subdomain or connected custom domain where pages go live without engineering tickets | Marketing lead + whoever controls DNS |
| Baseline measurement set | Impressions, clicks, average position, and enquiries or conversions for the last 90 days | Analyst or whoever owns GA4/Search Console |
| One named approver | A single person with authority to block a page. Not a committee | Marketing lead |
The baseline matters more than it sounds. Without a 90-day pre-period you cannot tell whether a new page cluster did anything. If your analytics are messy, spend week one fixing that rather than publishing.
If no single person can say "no" to a page, you do not have a gate - you have a suggestion box. The gate only works when one name is attached to the approval.
The 30-Day First Cycle: A Six-Step Routine
Here is the full arc, then we walk through the first four steps in detail; steps five and six follow in the second half.
- Days 1-3: Audit which domains AI systems currently cite in your category.
- Days 4-7: Define the query clusters you can answer credibly.
- Days 8-10: Write and circulate the publish gate.
- Days 11-20: Build a pilot batch, one intent per URL.
- Days 21-24: Publish in a controlled batch.
- Days 25-30 onward: Measure, then revise the same URLs.
Thirty days gets you one full loop and a pilot batch live. It does not get you results - search visibility compounds over months. The point of the first cycle is to prove the routine works so months two through six are boring and repeatable.
Step 1: Audit What AI Systems Already Cite in Your Category
Before you write anything, find out who currently wins the answer. Pick 10-15 realistic buyer questions in your category - the kind a customer would actually type, not the kind a keyword tool suggests. Ask each one in Google (check whether an AI Overview appears and which sources it links to), in ChatGPT, and in Perplexity. Record the cited domains in a simple sheet.
Query | Surface | Domains cited | Any ours? | Type of source
--------------------------------------------------------------
"bakkie for towing caravan" | AIO | 3 news/review sites | No | editorial
"family SUV under R450k" | ChatGPT | 2 dealer sites, 1 marketplace | No | commercial
"is diesel still worth it 2025" | Perplexity | 4 forums, 1 blog | No | editorial/UGCTwo patterns usually emerge. First, AI answers lean heavily on editorial, forum, and marketplace sources rather than brand pages. Second, most of your category competitors are absent too, which means the citation slot is genuinely open. This audit is job two of AI SEO, and most teams skip it entirely.
Run the same 15 questions again in 30 days. Citation share moves slowly, but seeing the same domains repeat tells you whether your cluster choices are realistic.
Step 2: Define the Query Clusters You Can Answer Credibly
Forget head terms. A small team wins by answering specific, low-competition questions where you hold genuine information - product specs, service availability, price bands, compatibility, location. Build clusters around four shapes:
- Use-case: "SUV for a family of five doing long-distance trips"
- Comparison: "diesel vs petrol double-cab for towing"
- Budget: "family SUV under R450,000"
- Specification: "tow rating explained for bakkies"
The filter is credibility. You can answer specification and budget queries if you hold the inventory data. You probably cannot credibly answer broad lifestyle questions that belong to publishers. Ecommerce teams can apply the same shapes to category clusters - see the ecommerce use case for how long-tail search coverage maps onto product ranges.
Step 3: Set the Publish Gate Before You Generate Anything
This is the step that separates a working AI SEO routine from a spam-policy incident. Google's spam policies explicitly address scaled content abuse - content generated at volume primarily to manipulate rankings - regardless of whether a human or a machine produced it. That is not a reason to avoid AI. It is a reason to write down, in advance, what every page must contain before it ships.
The gate should list five things: a stated fact source, a brand-voice check, structured data present, no thin or duplicated filler, and a named human approval logged with a date. Full detail is in the second half, but the principle is that rules written before generation are enforceable; rules written after publication are archaeology. Platforms built for this workflow, such as the Straider platform, front-load these checks rather than adding them as a cleanup step.
Step 4: Build Pages One Intent per URL, Grounded in Real Inventory or Service Data
Two rules do most of the work here. First, one intent per URL. A single page chasing "bakkie for towing", "family SUV under R450k", and "diesel vs petrol" will rank for none of them well, and it makes later revision impossible because you cannot tell which intent failed. Second, ground every page in real data: actual inventory, actual service coverage, actual prices. If you sell ten models, each model-plus-use-case page should name the model, its tow rating, its price band, and what it is genuinely suited to.
For a lean team, this is where AI-assisted drafting earns its place - not as a generator of whole articles from a keyword, but as a tool that takes structured input (model, specs, price, use-case) and produces a first draft you then check against the gate. The output of step four is not published pages. It is a staged batch waiting for step five.
