GEO / Roadmap / AI Citations

The 90-Day GEO Roadmap: How to Test, Measure and Grow AI Citations

Quick answer: A 90-day GEO roadmap is a delivery plan for improving and reviewing AI search visibility. Establish a baseline and page backlog, make a small group of documented improvements, then decide what to retain or revise. Measure citations, observed recommendations, visits and enquiries separately. The calendar does not guarantee when a platform will use a page.

A working calendar for a marketing team to establish a baseline, improve relevant pages and review source-specific AI visibility over a quarter. The schedule organises delivery; it does not promise when an engine will cite a page. Its companion, the AEO experimentation roadmap, covers question selection and measurement design.

By Carol Tan, Founder · Updated · 11 min read

On this page

  1. Week-by-week checklist
  2. What should you measure before day 1?
  3. Days 1–30: the foundation phase
  4. Days 31–60: the testing phase
  5. Days 61–90: the scaling phase
  6. Which KPIs matter, and which mislead?
  7. What AI Studio saw in its own data
  8. Frequently asked questions

Week-by-week checklist

The calendar is an example delivery sequence. Adjust it to access, evidence and review capacity; platform responses have no guaranteed deadline.

WeekPhaseDoMeasure
0BaselineRecord citations, cited pages, grounding queries; 10 buying questions; entity checkDay-0 snapshot saved
1FoundationCanonical entity description; Organization and Article schema validatedZero schema errors
2FoundationCheck relevant crawl controls, responses, canonicals and sitemap; review optional agent guidance where usefulAccess and current index evidence recorded; unresolved items assigned
3FoundationChoose buyer-decision clusters and a small group of existing owner pagesEach draft addresses a distinct information gap
4FoundationRelease reviewed changes; notify eligible updated URLs through the established process; start the weekly recordChanges and reporting coverage recorded
5TestHold the site stable; read first citations and grounding queriesWhich pages were retrieved, for what
6TestReview documented format or framing changes with a consistent question cohortAccess, index evidence and comparable observations recorded
7TestLaunch freshness and hub-and-spoke testsPer-page citations on the scoreboard
8TestReview complete periods and investigate missing data or weak page fitKeep, revise or investigate decisions with reasons; no automatic deletion
9ScaleExpand a useful approach only where another buyer decision warrants itEvidence and page ownership reviewed
10ScaleAddress a verified adjacent information gapAbsolute citations, relevant breadth and concentration reviewed together
11ScalePrepare relevant corroboration opportunities and contextual linksPermission, factual support and outreach scope reviewed; third-party publication remains independent
12–13ReportRe-run the 10 buying questions; write the day-90 report against the baselineCitations, cited pages, spread, hypotheses won

Takeaway: four weeks of engineering and writing, four of patience and reading, five of deliberate repetition.

What should you measure before day 1?

Start with a baseline you can compare with later complete reporting periods. The collection effort depends on available access, reporting history and the question sample. Preserve raw records, date ranges and missing-data notes.

Our guide to tracking AI citations across Bing, Search Console and GA4 covers each source in detail.

Days 1–30: what does the foundation phase include?

The first month has one job: make the site something an assistant can crawl, understand and attribute, then give it four pages worth citing.

Weeks 1–2: entity, schema and crawlability

Weeks 3–4: three clusters, four pages, one scoreboard

Keep a dated change log and avoid unnecessary simultaneous edits. Correct urgent errors promptly. A stable reference cohort may help interpretation, but neither a fixed waiting period nor an unchanged site isolates a change’s effect.

Days 31–60: how do you test what earns citations?

Month two is about hypotheses, not volume. You have four pages and a few weeks of citation data; the question is which properties of those pages the engines respond to.

Four hypotheses worth testing first

What a valid test looks like with noisy data

Citation data can move with demand, model behavior, source selection and reporting changes. Limit simultaneous variables where practical, define complete comparable windows and retain a stable reference cohort if suitable. These are observational checks unless the study design supports causal inference; three weeks is not a universal validity threshold. Use the AEO experimentation roadmap for the measurement procedure.

Reading the grounding-query report

Grounding queries describe the retrieval phrases returned by the report, not necessarily the original user question. Inspect the actual query-to-page association where available and check whether the page serves a relevant buyer decision. Unexpected associations merit review, not automatic reframing. A missing row or four weeks without a citation does not justify deleting a page; investigate access, coverage, usefulness and overlapping intent first.

Days 61–90: how do you scale what worked?

Month three doubles down on the framings that earned citations and deliberately spreads the risk.

Our GEO strategy guide for Singapore brands covers where Copilot, Gemini and AI Overviews behave differently.

Which KPIs matter, and which ones mislead?

Use source-labelled visibility measures alongside business outcomes. No single citation, page-count or referral metric establishes that a programme caused improvement.

Track attributable referral visits and received enquiries alongside citations. Attribution can be incomplete, so retain unknown sources rather than asserting a fixed delay or universal undercount. A citation is not a visit or a lead, and the two series should not be forced into a conversion rate. The AEO tools guide explains how different data sources support different decisions.

What AI Studio saw in its own data

Official Bing Webmaster Tools totals for aistudio.com.sg, re-summed 15 November 2026, include data through 12 November. 13 and 14 November are missing and are not zeros. Property citations were 6,742 in the seven days 27 October–2 November and 90 in the seven days 3–9 November. This page is in the 27 October–2 November page table with 951 citations and is absent from the later page table. An absent row is not a zero. URL Inspection on 15 November 2026 still showed this URL indexed, with crawl allowed and fetch successful. Ordinary Bing search for the property did not show the same break: 8 clicks and 277 impressions in 27 October–2 November, against 7 clicks and 324 impressions in 3–9 November. The cause is unproven. This note does not say citations have recovered, and the July reading above does not explain the November change.

A planning framework can organise work, but this incident does not show that publishing or markup changes caused a citation increase or decline. Review reporting definitions and the citation collection method before comparing windows.

Frequently Asked Questions

Is 90 days long enough to see AI citations move?

Ninety days is a planning and review period, not a guaranteed results window. Crawling, indexing, source selection and reporting can change on different schedules. Review complete source periods, page usefulness and business outcomes without promising a citation increase.

Do we need paid AI-visibility tools to run this roadmap?

Begin with the first-party reports available to the property and a clearly defined manual observation sample. Paid tools may help when their coverage, repeatability or exports fit a real need. Their value is a scope and capacity decision, not a universal day-61 milestone.

What if our citations are flat after the first 30 days?

First check reporting completeness, source scope, access and current index evidence. Then inspect actual query-to-page associations and the usefulness of the page for the buyer decision. Flat citations or missing rows do not establish a crawl or entity problem, and they do not create a guaranteed fix deadline.

How is this different from the AEO Experimentation Roadmap?

The AEO roadmap focuses on selecting questions, documenting changes and interpreting observations. This companion provides a possible delivery calendar for a wider GEO programme. Use both with source-specific metrics and business outcomes; neither establishes a controlled test or a promised citation lift on its own.

Can we run this alongside our existing SEO programme, or does it compete?

Coordinate both through shared page ownership, technical checks and a change log. Existing SEO work and useful content remain relevant to Google’s AI features. Avoid unnecessary simultaneous edits, but do not postpone urgent fixes merely to preserve a test window. Review visibility and business outcomes separately.

Should we measure traffic from ChatGPT and Perplexity instead of citations?

Measure both with their own definitions, plus received and qualified enquiries where available. A source citation does not imply a visit, and referral attribution can be incomplete. Do not assume a fixed lag or calculate a conversion rate across incompatible populations. Choose primary business measures according to the actual objective.

Related reading

Want to be the answer, not just a search result?

Featured in CNA

CNA interviewed founder Carol Tan about AI advertising on 24 August 2026. The coverage concerns creative production, not verification of search results.

Official profiles: AI Studio on Instagram · AI Studio on X · Carol Tan on LinkedIn.

Book an appointment WhatsApp Us