Buyer-question set
A versioned set of category, comparison, problem and purchase questions, split by market, language, topic and decision stage. Keep branded questions separate from open discovery.
90-day AI Search roadmap
A 90-day AI Search plan you can actually run.
Start with an agreed buyer-question set and a defensible baseline. Fix access and entity gaps before publishing. Build owned answers, strengthen the sources buyers already trust, then compare repeated observations. The dates organise the work. They do not promise when an AI engine will notice it.
Before day one
What has to be agreed before the clock starts?
The plan begins when the question set, source access and decision owners are ready. If those inputs arrive later, the dependent work moves with them.
Current sources
Website, product facts, approved claims, analytics, Search Console, CRM or booking data, and any AI visibility platform already in use. Unavailable sources stay unavailable.
Named owners
One business owner, one website owner, one content approver and one measurement owner. A person can hold more than one role, but every gate needs a name.
Decision rules
Agree what can ship without another meeting, what needs legal or brand approval, what makes an observation valid and when the question set can change.
Milestones and outcomes
What can the team control by day 30, 60 and 90?
Delivery milestones are commitments. Rankings, citations, recommendations, traffic and enquiries are observations controlled partly by engines, markets and buyers.
| Checkpoint | Controllable milestone | Engine-controlled observation |
|---|---|---|
| Day 30 | Accepted question set, technical and entity repair list, published high-priority fixes, measurement log and approved content queue. | Whether an engine recrawls a page, changes an answer, cites the business or sends a visit. |
| Day 60 | Priority answer pages published, proof gaps assigned, source-work priorities and approved angles ready, and two comparable reporting periods reviewed. | Whether a publisher accepts a contribution, a community discussion stays visible or an engine repeats the new evidence. |
| Day 90 | One complete delivery cycle, a versioned evidence trail, a keep, change or stop decision for each workstream and the next 90-day backlog. | Any promised position, recommendation, traffic level, enquiry volume or revenue result. |
01 Days 0 to 14
Accept the question set and establish the baseline
Make the measurement contract explicit before anyone optimises against it. The first useful output is a question set and a baseline people can challenge, not a dashboard screenshot.
Inputs required
- Sales and support language
- Current offer and approved claims
- Target markets, languages and competitors
- Available analytics, Search Console, CRM or booking sources
Owner
Business owner, measurement owner and the people closest to sales and customers.
Work
- Group questions by topic, market, language and decision stage.
- Separate branded recall from open-category discovery.
- Run the agreed questions across the named AI systems and retain the answer, date, market, visible sources and, where exposed, model, version or interface.
- Record missing, failed and unsupported observations. Do not convert them to zero.
- Define visibility, share of voice, position, source use, sentiment, accuracy and qualified outcomes before reporting them.
Evidence
- Versioned question register
- Prompt-level baseline with valid counts and missing-data log
- Source and competitor gap map
- Signed measurement definition and change log
02 Days 15 to 30
Repair access, entity clarity and the most expensive answer gaps
Fix the conditions that make useful work findable and believable. Start with blockers that affect many questions, then repair the pages closest to commercial intent.
Inputs required
- Technical crawl and index checks
- Organisation, product, location and author facts
- Prompt-level source gaps
- Approved website and CMS access
Owner
Website owner leads technical delivery. Business and content owners approve facts and claims.
Work
- Check crawl access, index eligibility, internal links, status codes, preferred-page signals and rendered text.
- Reconcile organisation, product, location and author facts across the site and trusted profiles.
- Make structured data match visible content. Treat it as description, not a visibility shortcut.
- Where agents need to complete a task, test navigation, forms, booking or commerce journeys through the rendered page and accessibility tree.
- Rewrite the highest-value pages so each section answers a real buyer question clearly and supports claims with current evidence.
- Fix broken destinations and instrument the roadmap-to-audit and roadmap-to-managed-service paths without collecting personal data in analytics labels.
Evidence
- Before and after technical checks
- Entity and claim register with source dates
- Published-page and structured-data readback
- Accessibility, route, analytics and browser evidence
03 Days 31 to 60
Build owned answer coverage and close proof gaps
Publish a small number of strong pages that answer missing buyer questions. Add evidence, examples and expert ownership that another source can verify.
Inputs required
- Accepted content priorities
- Subject-matter interviews or first-hand evidence
- Approved claims and product facts
- Publishing capacity and review dates
Owner
Content owner drafts, subject expert verifies, website owner publishes and measurement owner records the change.
Work
- Choose answer pages from prompt and source gaps, not a generic keyword list.
- Give each page one clear job and make every major section useful in isolation.
- Use first-hand detail, original data or a named method where it genuinely exists.
- Connect each new page to the offer, relevant evidence, the audit and the measurement method.
- Review the first complete post-change period by engine, topic, market and language before changing direction.
Evidence
- Content brief tied to target questions
- Source ledger and expert approval
- Published locale-native pages with link and schema checks
- First complete post-change comparison with unchanged definitions
04 Days 61 to 90
Strengthen cited sources, repeat measurement and decide what earns another cycle
Owned pages are only one part of the evidence system. Work on relevant independent sources, then use repeated observation to choose the next cycle.
Inputs required
- Source-gap map refreshed from actual answers
- Approved expert angles and evidence
- Publisher or community rules
- At least two comparable completed periods where volume allows
Owner
Subject expert owns the contribution. Outreach or community owner manages the relationship. Measurement owner keeps interpretation separate from delivery.
Work
- Prioritise publications, directories, reviews and public discussions already used for the target questions.
- Offer useful expertise, data or corrections that stand without a sales pitch.
- Follow editorial and community rules. Publication and moderation decisions remain with the platform owner.
- Compare completed periods with the same definitions and show absolute counts beside rates.
- For each workstream, decide to keep, change, pause or stop based on delivery evidence, answer movement and commercial observation.
Evidence
- Outreach and contribution ledger with statuses
- Published or declined source outcomes
- Repeated engine, topic and market comparison
- Roadmap-to-audit and roadmap-to-managed-service transitions, completed audits and qualified booked demos where authoritative sources are available
- Next-cycle backlog with explicit reasons
Measurement cadence
What gets measured during the 90 days?
Run the agreed questions on a consistent schedule, review complete periods and keep business observations in a separate funnel. A changed answer is not automatically a commercial result.
- 01
Daily or scheduled
Prompt responses, brand mentions, visible order, source links, failed runs and changed facts. Use the actual platform cadence and label it.
- 02
Weekly
Completed-period movement by engine, topic, market and language. Review source gaps and factual errors before changing the content queue.
- 03
After each release
Record the page, claim, source, date, approval and verification. Mark the intervention on the measurement timeline.
- 04
At days 30, 60 and 90
Compare like with like. Show counts, missing coverage and material method changes. Decide what to keep, change, pause or stop.
- 05
Commercial observation
Measure privacy-safe transitions from this roadmap to the audit and Managed AI Search pages, completed audit requests and qualified booked demos downstream. Current authoritative baselines are unavailable, so the first post-release read establishes them.
DIY, tool or managed service
Who should run the plan?
The work needs ownership more than it needs a particular vendor. Choose the lightest operating model that can maintain the evidence and shipping cadence.
| Route | Best fit | What you still own | Main trade-off |
|---|---|---|---|
| DIY with an internal owner | A focused team with website access, subject expertise and enough time to run the same questions repeatedly. | Question acceptance, publishing, source work, QA, measurement definitions and the 30, 60 and 90-day decisions. | Lowest external spend, highest demand on internal judgement and delivery time. |
| AI visibility tool plus internal team | A team that can execute but needs repeatable prompt, competitor and source data. | Prompt quality, interpretation, technical changes, content, outreach and commercial attribution. | Better measurement capacity does not ship the work or guarantee an engine outcome. |
| Specialist or agency support | A team that needs focused research, technical or editorial capacity while retaining programme ownership. | Approvals, source access, product truth, implementation access and final decisions. | Useful for a bounded gap, but handoffs can slow a plan if ownership is fragmented. |
| Schmitdy Managed AI Search | A business that wants one team to run research, page work, source work, verification and reporting in its working channel. | Product truth, timely access and approvals, legal or brand decisions and commercial qualification. | Higher service cost, with more of the operating burden handled. Engine and publisher outcomes remain outside Schmitdy control. |
Ownership
What does each owner decide?
Business owner
Accepts target markets, priority questions, approved claims, commercial qualification and the day 30, 60 and 90 decisions.
Measurement owner
Versions the question set and definitions, records unavailable data, checks comparability and reports uncertainty.
Website owner
Controls crawl, index, CMS, analytics and release quality, and keeps visible copy aligned with structured data.
Content and expert owner
Provides first-hand evidence, approves factual language and rejects weak or duplicated pages.
Source-work owner
Manages publisher and community relationships, follows rules and records acceptances, declines and removals without treating them as failures of the whole plan.
Public sources
Sources and limits behind this roadmap
These sources explain platform mechanics and measurement. The sequence above is Schmitdy’s operating synthesis, not a timetable endorsed by an AI engine.
- Google Search Central: AI features and your websiteExisting SEO foundations still apply. Eligibility, crawling, indexing and serving are not guaranteed. Search Console now has a Generative AI performance report whose data comes from the Web search type.
- Google Search Central: optimising for generative AI featuresGoogle prioritises useful, original, people-first content and warns against commodity content, scaled pages and supposed AI-search hacks.
- Google Search Console: Generative AI performance reportThe report covers impressions from AI Overviews and AI Mode. Low-volume data may be unavailable, and exports turn unavailable values into zero, so retain the missing-data record.
- OpenAI: Publishers and Developers FAQOAI-SearchBot access affects discovery for summaries and snippets, and ChatGPT search referral links can include utm_source=chatgpt.com.
- Peec AI: understanding your performancePrompt runs are the measurement unit, and visibility, competitors, sources and recent answers can be reviewed separately.
FAQ
Questions to settle before you start
Can this roadmap guarantee a ChatGPT recommendation by day 90?
No. The roadmap can guarantee only the work you control and verify. AI systems decide when to crawl, retrieve, cite, mention, order or recommend a business.
How many buyer questions should we track?
Use the smallest set that covers the real category, comparison, problem and purchase decisions across the required markets and languages. As a worked starting frame, try five to ten questions in each of those four groups for one market and language, then remove any cell the team cannot repeat consistently.
Do we need an AI visibility platform?
No. A team can retain answers and sources manually for a bounded set. A platform becomes useful when repeated multi-engine, multi-market tracking would otherwise consume the team’s delivery time.
What if analytics, Search Console or booking data is unavailable?
Record the source as unavailable. Do not write zero and do not infer that no visits, requests or bookings occurred. Establish the baseline after access is restored and state the observation window.
Should technical work wait until the full baseline is complete?
Fix clear access failures and broken user paths immediately. Hold broad content expansion until the question set and source gaps are accepted, otherwise the team can publish a lot without answering the right demand.
How often should the plan change?
Review priorities weekly, but change the measurement contract only with a new version. Major question, engine, model, interface, market or weighting changes break direct comparison and need an annotation or fresh baseline.
What is the difference between a milestone and an outcome?
A milestone is delivered work with evidence, such as a repaired route or published answer page. An outcome is observed behaviour, such as an engine citation, a visit or a qualified demo. The team controls the first and can only influence the second.
What happens after day 90?
Use the evidence to keep, change, pause or stop each workstream. The next cycle should expand only the questions, pages and sources that still matter commercially and can be measured consistently.
Your website stack
Keep your platform. Work through real publishing paths.
Next step
Start with the questions, the evidence and the owner.
The free audit gives you a bounded starting point. If you want Schmitdy to run the research, shipping and measurement with your team, compare the managed service before you choose.