Workflow & routines
GEO for marketers: a weekly routine for improving AI search visibility
Run GEO as a repeated operating routine, not a one-off check. Keep prompts and aliases stable, review repeated samples, inspect cited sources, then request briefs for the largest gaps. Later scans are observations, not proof that a published change caused the movement.
GEO is not a one-off project
AI answers vary between runs and can also change as models, indexes and published material change. There is no universal weekly movement rule, but a stable operating cadence makes those changes observable. Paid Suparanku plans can schedule recurring scans; Free provides one initial audit.
Setup: build the measurement base (first week only)
Turn customer questions into prompts
Collect the questions buyers actually ask AI — not keywords. Sales call notes, search queries, FAQs and deal memos are your raw material. Prioritize purchase-stage questions: “what’s the best X?”, “how do A and B compare?”, “what does X typically cost?”.
Register spelling variants
Brand mentions vary in spelling — especially in Japanese, across katakana, romaji, abbreviations and spacing. Without registered variants, real mentions get counted as zero. Capture every name customers actually use, not just the official one.
The weekly routine: 30–60 minutes
1. Read the deltas, not the dashboard (10 min)
Start with what changed since the previous comparable scan: prompts with a visibility movement, changes in answer position, and sentiment shifts. Keep the prompt set and engines stable when comparing periods. Repeated runs are necessary, but small deltas from only a few samples remain directional rather than statistically significant.
2. Check cited sources (10 min)
Review the domains and articles AI newly cites. If an outdated third-party post is being treated as your official information, that is your top fix. Also note which of your own pages get cited — that reveals the content shape AI prefers.
3. Convert opportunities into briefs (20 min)
“Only competitors are recommended for this question” and “we have no useful page on this topic” are hypotheses to investigate. When a content gap is supported by the evidence, request a brief that specifies what to write, where it belongs and which format fits. The KDD 2024 study reported gains of up to about 40% for documents already present in a fixed context; it did not prove discoverability, traffic or growth. Statistics, named sources and relevant quotations are options to use where they improve the piece, not three mandatory fields in every brief.
4. Re-measure after publishing (10 min)
Record the published URL and keep measuring the relevant prompt set. Indexing, retrieval and citation lag varies by platform and page; there is no fixed “few weeks” causal rule. A later appearance is useful observational evidence, but it does not by itself prove that the publication caused the change.
What not to do
- Unsupported “pick us” copy — replace self-assertion with verifiable facts, useful comparisons and clear sourcing.
- Recycled keyword stuffing — the Princeton experiment put keyword stuffing below its unoptimized baseline, and Google says conventional Search fundamentals apply to its generative features.
- Changing the measurement set mid-comparison — adding prompts or engines changes the denominator. Record the change and start a new comparable baseline.
Suparanku automates scans, comparisons and prioritized recommendations. Content briefs are generated on demand and consume the plan’s monthly quota; Free has no monthly content-brief allowance. Publishing remains the work of the customer, agency or authorized agent, and tracked URLs connect that work to later measurement.
* Aggarwal et al., KDD 2024 (10,000-query experimental comparison; fixed-context scope) ** Google Search Central, “AI features and your website”