AI Search Optimization Roadmap: Your 90-Day Plan to Get Cited by AI

Why the Plan Runs for 90 Days

AI search optimization spans technical access, useful visible content, source quality, and repeated measurement. Public provider documentation does not establish that completing every layer will produce a citation. This plan covers both GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization). Track each layer separately so an access fix, page change, or independent mention is not mistaken for a citation outcome.

A 90-day window gives you an operating cadence without promising an outcome. It gives you time to establish a baseline, make one controlled change at a time, observe the same prompt-engine panel, and compare results by engine. Crawl requests, indexing, citations, referrals, and conversions should remain separate measurements.

Some changes may appear quickly in one engine and remain absent in another. Do not infer progress from a single answer. Use repeated samples, preserve zero-result cells, and report uncertainty when the panel is small.

The plan is divided into six two-week sprints, each focused on a specific layer of optimization. You can run some work in parallel, but keep measurement windows and change logs intact. By the end, you will have a repeatable audit, implementation, monitoring, and review system.

Here is the roadmap at a glance:

• Weeks 1-2: Baseline audit. Understand where you stand today.
• Weeks 3-4: Technical foundation. Make your site AI-accessible.
• Weeks 5-6: Content optimization. Rewrite and restructure for AI citation.
• Weeks 7-8: Source-layer work. Earn accurate independent mentions and publish reproducible evidence.
• Weeks 9-10: Monitoring and iteration. Measure results and refine.
• Weeks 11-12: Scale and systematize. Build workflows that preserve measurement and ownership.

Each phase has a reviewable deliverable. Set the weekly time box from the number of pages and prompts in scope. Technical owners can handle access and markup, content owners can handle answer pages, and marketing owners can handle independent sources and research.

What You Will Need

Before starting, gather these resources:

• A spreadsheet tool (Google Sheets, Excel, or Notion) for tracking your baseline audit and ongoing metrics
• Access to your website's CMS for making content changes, adding schema markup, and updating robots.txt
• Accounts on all 5 AI platforms (ChatGPT, Perplexity, Claude, Gemini, Google). Free tiers are sufficient for testing.
• Google Search Console access for sitemap submission and crawl monitoring
• A weekly time box sized to the prompt panel and number of priority pages

You can execute the measurement loop manually. Foglift's AI Visibility platform stores prompt runs, mentions, citations, sentiment, and competitors so the same panel can be compared over time.

Let us walk through each phase in detail.

Weeks 1-2: Baseline Audit

Start with a dated baseline. It gives each later observation a fixed prompt, engine, page, and denominator to compare. Without that record, a changed answer cannot be separated from a changed test setup.

The baseline must preserve prompt wording, engine, date, answer text, cited URLs, and zero-result cells. Without those fields, a later change cannot be compared cleanly with the starting point.

Think of the audit as your diagnostic scan before treatment. A doctor does not prescribe medication before understanding the patient's symptoms. Similarly, you should not start implementing technical fixes or content changes until you have a clear, data-backed picture of your current AI search visibility.

Week 1: Run Your Technical Audit

Start by running your domain through Foglift's free Technical Audit. This gives you an instant snapshot of your technical readiness: schema markup, AI crawler access, content structure, and overall AI Readiness Score. Save this report. It is your day-one benchmark.

Then run the free AI Brand Checker for the separate outcome baseline. It tests buyer-shaped prompts across five answer surfaces and records brand mentions, competitors, citations, and a prioritized action plan without requiring signup.

Next, manually query each of the five major AI engines. Prepare 5-8 queries your ideal customers would ask:

• Brand queries: "What is [Your Brand]?" and "Is [Your Brand] legit?"
• Category queries: "Best [your product type] in 2026" and "Top [category] tools"
• Problem queries: "How to solve [problem you fix]" and "Why is [pain point] happening?"
• Comparison queries: "[Your Brand] vs [Competitor]" and "[Competitor] alternatives"

Run every query on ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews. Use incognito mode. For each, record: whether your brand was mentioned, the exact wording, the sentiment (positive, neutral, negative), and which competitors appeared instead.

Create a spreadsheet with columns: Query, Platform, Brand Mentioned (Y/N), Sentiment, Competitors Listed, Notes. This document becomes your ongoing tracking tool for the entire 90-day plan.

Save the exact wording used for your brand and competitors. A phrase such as "a newer player" is an observation to compare across runs. A factual error should be traced to the cited source before you decide whether to correct an owned page, an external profile, or another source.

Also note the format of responses. Do AI engines list your competitors with bullet points and descriptions while your brand gets no mention at all? Or does your brand appear but without a link or detailed context? The level of citation matters: a passing mention is very different from being the recommended solution with a linked source. Categorize each mention into one of four tiers:

1. Tier 1, Primary recommendation: AI explicitly recommends your brand as the best or top solution
2. Tier 2, Named in a list: Your brand appears in a list of options with a brief description
3. Tier 3, Passing mention: Your brand is mentioned but not highlighted or recommended
4. Tier 4, Absent: Your brand does not appear at all in the response

Use the tiers as observation labels, not a causal ladder. A move from absence to a passing mention is a changed outcome for that sample. It does not prove that a model has learned the brand or that a later answer will rank it higher.

Week 2: Document Current Visibility Across All 5 Engines

With your raw data collected, calculate your baseline metrics:

• Presence rate: Total mentions divided by total query-platform combinations. If you ran 8 queries across 5 platforms (40 checks) and appeared in 6, your presence rate is 15%.
• Platform breakdown: Which engines mention the brand, and which cited sources support the answer? Report each engine separately.
• Sentiment score: Of your mentions, what percentage are positive versus neutral or negative?
• Accuracy rate: Are AI engines saying correct things about your brand? Flag any inaccurate claims (outdated pricing, wrong product descriptions, incorrect comparisons).
• Competitor gap: How many times do competitors appear where you do not? Identify the top 3 competitors who show up most frequently.

Also audit your technical foundation using the AI search audit guide. Document any blocked crawlers, missing schema, or content structure issues. These become your priority fixes for weeks 3-4.

Create a one-page "AI Visibility Snapshot" that summarizes your current state. Include your presence rate, the three AI engines where you perform best, the three where you are weakest, your top competitor threats, and the critical technical issues found in your Technical Audit. This snapshot becomes the reference document you will compare against at weeks 6, 9, and 12.

By the end of week 2, you should have a complete visibility report, a list of technical gaps, a source-gap brief, and a change ledger. Set process targets you control, such as repairing a false fact, publishing a documented answer, or earning review from one relevant independent source. Do not promise a presence-rate target before you have a measured baseline and learning window.

Weeks 3-4: Technical Foundation

This phase fixes infrastructure that affects whether search crawlers can request your content and whether machines can extract explicit page information. Treat crawler access, indexing, and citation as separate checks. Technical readiness is necessary evidence, but it does not guarantee an AI answer will use the page.

Match crawler controls to provider roles. OpenAI documents OAI-SearchBot for search and GPTBot for potential training use. Anthropic documents Claude-SearchBot, ClaudeBot, and Claude-User separately. An allow rule removes one access barrier; it does not prove indexing, retrieval, or citation. Google also says no special AI schema is required for its AI features.

Week 3: Schema Markup and Structured Data

Schema markup provides explicit machine-readable labels for information already visible on the page. It can reduce ambiguity around your organization, articles, products, and authors. It does not guarantee retrieval, ranking, or citation, and it should never replace clear visible copy.

JSON-LD adds machine-readable information about visible page content. Use the implementation method supported by your site and validate the rendered result. Google says structured data can support eligibility for documented Search features, but correct markup does not guarantee that a feature will appear.

Choose schema by page type and Google's current feature documentation. Do not add every type to every page. For a comprehensive walkthrough, see the schema markup for AI search guide:

• Organization: Use applicable identity fields on the organization or home page.
• Article: Use it for an article and keep headline, author, and dates synchronized with visible copy.
• FAQPage: Use it only when the page and Google's current eligibility rules apply, and keep every answer visible.
• Product: Follow Google's product documentation for an eligible product page and current offer data.
• LocalBusiness: Use the most specific applicable subtype on a page that describes the real location.

Validate every schema implementation using Google's Rich Results Test. Common mistakes include missing required fields, malformed JSON, and schema that does not match the visible page content. Fix all validation errors before moving on.

More markup is not inherently better. Google's policies require structured data to represent the main visible content and follow the specific feature guidelines. Remove irrelevant, stale, hidden, or unsupported fields.

By the end of week 3, inventory the markup already present, repair invalid or stale records, and add an applicable type only where it describes visible page content. A page can be useful and eligible for Google's AI features without special schema.

Week 4: Robots.txt, Sitemaps, and Site Architecture

Crawler policy, sitemap discovery, internal links, indexing, and answer inclusion are separate controls. Review each one without treating access as proof of a later citation.

Decide which provider roles you intend to allow. This week, compare your policy with current first-party crawler documentation:

• OAI-SearchBot (ChatGPT search): allow if you want ChatGPT search discovery
• Claude-SearchBot (Claude search): allow if you want Claude search discovery
• PerplexityBot (Perplexity search): allow if you want Perplexity indexing
• Googlebot (Google Search and AI Overviews): keep important public pages crawlable
• Google-Extended: decide separately because it controls Gemini Apps and Vertex AI model training and grounding, without affecting Google Search

A robots.txt edit may be only one part of the fix. CDN, WAF, authentication, and origin rules can still deny a request. See the robots.txt for AI crawlers guide for current provider roles and verification steps.

A common gotcha: some CMS security plugins and CDN configurations add AI crawler blocks by default. Check your robots.txt, server-level access rules, Cloudflare bot management settings, and CDN configurations. Cloudflare's "Bot Fight Mode" and similar features can inadvertently block legitimate AI crawlers. If your server logs show that GPTBot or ClaudeBot is getting 403 errors, the block may be happening at the infrastructure level rather than in robots.txt.

Next, review your XML sitemap. Include the canonical public pages you want search systems to discover and use accurate lastmod dates only when page content materially changes. Reference the sitemap in robots.txt and submit it through Google Search Console.

Finally, review internal discovery. Ensure every priority page has at least one crawlable internal link from a relevant page and is present in the intended navigation or sitemap. Do not publish a universal click-depth threshold as a citation rule; providers do not document one.

Draw a simple graph of the priority pages and their internal links. Flag orphaned pages, duplicate paths, stale canonicals, and important pages omitted from the sitemap. Repair the discovery path, then verify it with a crawl rather than assuming an AI engine used it.

Week 4 deliverables: AI crawlers unblocked in robots.txt, updated XML sitemap with accurate lastmod dates, site architecture audit complete, orphan pages linked, and internal linking improved across your top 20 pages. Run another Technical Audit at the end of this week to see if your AI Readiness Score has improved from your week 1 baseline.

Providers do not publish a universal reflection schedule for these changes. Keep the prompt panel fixed, annotate the release date, and compare repeated observations without claiming that one engine should move before another.

Weeks 5-6: Content Optimization

With access policy documented, compare your priority pages with the pages actually cited for each prompt. Improve missing facts, direct answers, source links, and page structure where the benchmark shows a concrete gap. For the full playbook, review the AI-friendly content architecture guide.

Search-enabled answers can quote or summarize web sources. Write important claims so they remain accurate when read outside the surrounding paragraph, and attach a direct source to claims that depend on external evidence.

Do not impose a universal sentence count. Use enough context to state the subject, fact, boundary, and source clearly. Then inspect real citations to see which passage and page an engine selected.

Week 5: Rewrite Top Pages for AI Citation

Identify your 5-10 most important pages: the ones that correspond to the queries you tested in your baseline audit. These are typically your homepage, main product or service pages, pricing page, and top 3-5 blog posts that address your core topics. For each page, apply these optimization principles:

• Lead with the answer: Put the direct, bounded answer near the heading so readers can find it quickly.
• Use descriptive headings: Phrase headings around the reader's job. A question is useful when the section genuinely answers it.
• Add verifiable facts: Include current prices, plan boundaries, dates, methods, and direct sources where they help the decision.
• Create self-contained paragraphs: Each paragraph should make sense when extracted from the page and quoted on its own. Avoid "as mentioned above" references that break when taken out of context.
• Include comparison context: Compare the same dimensions and link to each vendor's current first-party source.

Keep the page useful for its human reader. Use descriptive headings, visible evidence, and self-contained explanations so the answer and its source remain clear when a passage is extracted.

Replace vague copy with a fact you can defend. For example: "[Product Name] includes async updates, shared workspaces, and automated status summaries on the Launch plan at [current price]." Add outcome numbers only when the page discloses the sample, date, method, and relevant limitations.

Prioritize page rewrites from the baseline audit. Start with pages tied to decision-relevant queries, then compare each page with the sources actually cited for that prompt. Limit week 5 to a small page set so each change is reviewable and the next measurement window still has a useful change ledger.

A useful framework for prioritizing which pages to rewrite first: rank each page on two dimensions, business value (how important is the associated query to your revenue?) and citation potential (how likely is this page to be cited if properly optimized?). Pages with both high business value and high citation potential are your week 5 priorities. Pages with high business value but low citation potential may need more fundamental restructuring and can be queued for later.

Week 6: Add FAQ Sections and Entity Pages

FAQ sections are useful when they answer real customer questions that the main page does not already resolve. No public cross-engine study establishes a citation lift from FAQPage markup alone. Google says no special schema is required for its AI features, and valid structured data does not guarantee a Search feature.

Add an FAQ only where it improves the page. Use real questions from support, sales, onboarding, and search evidence. Keep the answer long enough to be accurate and short enough to scan.

Where do you find the right FAQ questions? Start with these sources: your customer support inbox (what do people actually ask?), your sales team's notes (what objections and questions come up in discovery calls?), Google's "People Also Ask" section for your key queries, and the AI engines themselves. Ask them questions about your category and note what follow-up questions they suggest.

Improve an existing canonical definition before creating another page. A useful definition identifies the term, scope, examples, source, and related concepts without duplicating a stronger route.

A useful definition page gives readers and retrieval systems one canonical record for a concept. Include the scope, examples, primary sources, related terms, author, and update date. Measure whether the page enters citation sets instead of assuming that a glossary page creates authority.

Practical checklist for this week:

• Add FAQs only where customer questions remain unanswered
• Add FAQPage JSON-LD only where current eligibility and visible-content rules apply
• Improve an existing canonical definition page before creating a new one
• Cross-link entity pages to and from your main content pages
• Check that all new content passes the "extraction test": can each paragraph stand alone as a citation?
• Verify all new schema validates without errors using Google's Rich Results Test
• Update dateModified in your Article schema for every page you updated

At the end of week 6, re-run the fixed panel and document the observation. Do not assign a faster reflection schedule to one engine without measured evidence. A changed answer is a signal for another sample, not proof of causation.

Weeks 7-8: Authority Building

This phase moves from owned pages to the source layer observed in the baseline. Classify winning sources as independent editorial, community, reference, or vendor-controlled. Pursue accurate inclusion only where the source and relationship are legitimate.

Start with the sources in the baseline answers. An independent review, community thread, provider document, and competing vendor page have different ownership and editorial boundaries. Record which source types recur by engine before choosing a distribution target.

Treat "entity authority" as an association to measure, not a score you can claim from publication alone. Keep product facts consistent, publish methods with original data, and seek accurate independent corroboration. Then test whether those sources appear for the fixed prompts.

Week 7: Earn Citations and Build Backlink Authority

Independent coverage can corroborate first-party claims and can itself become a cited source. Do not claim that a backlink causes an AI model to trust, prioritize, or learn the brand. Measure whether the independent page enters the citation set after publication.

Choose source-layer opportunities that fit the winning-source benchmark:

• Contribute to industry publications: Offer expert commentary, reproducible data, or a correction when an editor already covers the topic.
• Review relevant directories: Correct inaccurate product facts and maintain categories that match the product.
• Respond to journalist queries: Provide a direct source, disclosed method, and bounded quote that can be checked independently.
• Create a sourceable asset: Publish original research, a maintained guide, a free tool, or a data visualization with a clear method and update date.

Track each independent mention by URL, publication date, claim, disclosure, and target prompt. The ledger supports later measurement without promising a model-update effect.

Set a week 7 goal from the benchmark rather than an arbitrary mention count. Track the URL, source type, editorial relationship, publication date, claim, link status, and target prompt. Independent editorial coverage and self-submitted directory records should remain separate rows because they are different evidence.

A newsroom can maintain accurate links to independent coverage, awards, and external references. Include only real mentions, disclose sponsored relationships, and respect logo and quotation permissions. The page is a source record; its publication does not prove that an engine will retrieve or use it.

Week 8: Create Original Data and Publish Research

Original data can create a unique primary source when the method, sample, limitations, and downloadable evidence are clear. Publication does not force an engine to retrieve or cite it.

Ways to create original data this week:

• Customer survey: Disclose recruitment, question wording, response count, field dates, and limitations.
• Internal data analysis: Aggregate anonymized data from your product usage into industry benchmarks. "We analyzed 10,000 [actions] and found that..."
• Market research: Compile publicly available data into a comprehensive industry report with your analysis and commentary.
• Case studies: Document real customer outcomes with the baseline, denominator, date range, method, and material limitations.
• Benchmark reports: Test and compare products, tools, or approaches in your industry. Publish the results as a data-driven comparison.

Publish your research as a dedicated page with clear data points, charts or tables, methodology notes, and a publication date. Add Article schema with the author and datePublished fields. Make the key findings easy to extract: use numbered lists, bold key statistics, and provide a summary at the top.

Lead with a summary of three to five findings, then provide the full analysis, methodology, limitations, and detailed breakdowns. Include a publication date and state an update cadence only when you can maintain it. This structure makes the evidence easier to inspect and quote. Retrieval and citation still require measurement.

Once the research is published, distribute it to relevant communities and publications with the method, dataset, and key finding attached. Log each event and inspect later answers for the report URL. Do not infer model trust from link count or promise that one format will outperform another.

By the end of week 8, complete one well-documented source-layer intervention and log every distribution event. The prompt panel may still show no movement. Preserve that result and wait for the planned learning window before choosing the next action.

Weeks 9-10: Monitoring and Iteration

You are now past the halfway point. Do not assume the earlier changes have been crawled or reflected in answers. This phase checks server evidence, index state where available, and the fixed prompt-engine panel.

Treat each intervention as a hypothesis. A page-level change and a later citation can be correlated in a small panel without proving that the change caused the answer. Preserve control prompts, zero-result cells, and any concurrent changes.

Weeks 9 and 10 are the review phase. Compare the same evidence fields, note confounders, and choose the next action from the largest repeated gap rather than the most flattering sample.

Week 9: Set Up AI Visibility Monitoring and Re-Run Your Audit

Re-run the exact queries from week 1 with the same wording and engine set. Add columns for the new answer, citation set, date, and model. The comparison shows outcome movement in the observed panel; it does not isolate impact when several changes shipped together.

Calculate your updated metrics:

• Presence rate change: Report the numerator, denominator, engine, and date range with the percentage.
• New engine coverage: Record where the brand newly appears without assigning a universal reflection speed.
• Sentiment shift: Are AI engines now describing your brand more accurately and positively?
• Competitor movement: Have your competitors also improved, or have you gained ground?

This is also the week to set up automated AI Visibility monitoring so you no longer need to run manual checks. Foglift's monitoring workflow lets you define the queries that matter to your business and track your brand's presence across all five AI engines automatically. You receive weekly reports showing where you appear, what the AI says about you, and how your visibility trends over time.

If you are not ready for paid monitoring, set a calendar reminder to manually re-run your audit queries every two weeks. Consistency in tracking matters more than the tool you use.

Keep the original panel frozen for comparison. Add newly discovered prompts to a separate expansion panel so the denominator and history remain interpretable.

Week 10: Track Score Changes and A/B Test Approaches

With your week 9 audit data in hand, compare each repeated query and engine with its week 1 baseline. Record the eligible runs, changed pages, release dates, and concurrent interventions beside every result. This evidence can select the next test, but a before-and-after movement alone does not identify its cause.

Identify which results justify a follow-up test. Start with these diagnostic patterns:

• Markup correlation: If marked-up pages moved, verify visible-content changes and other confounders before expanding the change.
• Format correlation: If one page type appears more often, compare prompt intent, source authority, freshness, and sample size.
• Engine divergence: Different outcomes do not prove one engine crawled faster or another used training data.
• Prompt divergence: Inspect the winning cited pages before applying a pattern elsewhere.

Run controlled tests only when pages, prompts, and source authority are comparable. Pre-register the change and observation window. Treat the result as directional unless the repeated sample is large enough to support a stronger inference.

Other useful tests to run during week 10:

• Pages with data points and statistics versus pages with general advice: which gets cited more?
• Complete answer coverage versus concise answer coverage for the same intent
• Pages with comparison tables versus pages with narrative comparisons
• Content with explicit source citations versus content without them

Document the results of every test in a dedicated "GEO Experiments" sheet in your tracking spreadsheet. For each test, record: what you tested, the hypothesis, the pages involved, the measurement period, and the result. Over time, this becomes an invaluable knowledge base for your organization. New team members can review past experiments to understand what works in your specific market without re-running the same tests.

Specific, sourced claims are easier to verify and maintain than vague copy. If one page earns three citations and another earns none, record the difference and inspect confounders. Do not generalize from that pair alone.

Also review your AI search optimization checklist score. Compare every category with the week 1 assessment and label it improved, unchanged, or lower. Use the evidence behind the largest repeatable gap to choose the final sprint rather than predicting that most categories will improve.

If an engine does not move, inspect its cited sources and current first-party crawler documentation. Avoid undocumented claims about training snapshots, top-10 dependence, or which engine responds fastest.

Here is a platform-specific quick guide for troubleshooting:

• Perplexity not citing you: Verify PerplexityBot policy and fetches, then inspect the sources it cites for the exact prompt.
• ChatGPT not mentioning you: Verify OAI-SearchBot policy and inspect the cited sources in a search-enabled answer.
• Google AI Overviews excluding you: Check Googlebot access, index and snippet eligibility, Search Console evidence, and the sources shown for the query.
• Claude not referencing you: Verify Claude-SearchBot policy and inspect the current cited-source set.
• Gemini Apps not grounding on your pages: Check Google-Extended separately. Google says it controls Gemini Apps and Vertex AI grounding but does not affect Google Search.

Weeks 11-12: Scale and Systematize

The final phase turns the first 10 weeks into an owned operating process. Document the panel, change ledger, technical checks, source review, and decision rules so another teammate can reproduce the next cycle.

Write down which evidence triggers a page edit, source-layer action, technical escalation, or no-change decision. Assign an owner and review date to each recurring task.

A documented SOP, a monitoring record, and explicit ownership reduce ambiguity when the next result changes. They also make it easier to distinguish a real intervention from routine publishing activity.

Week 11: Create Workflows and SOPs

A process held by one person is difficult to reproduce or audit. Document the procedures developed during the past 10 weeks so another owner can run the same checks:

• New content workflow: Require a direct answer, primary sources, accurate metadata, and only the schema types supported by the visible page and current eligibility rules.
• Content refresh workflow: Review material facts on a defined cadence. Change dateModified only when the page itself materially changes.
• AI search audit workflow: Define who runs the audit, which fixed queries and engines are tested, and where results are stored. Size the cadence to the panel and use a structured audit process.
• Source-layer workflow: Record eligible independent sources, the offered evidence, editorial relationship, and later citation observation.
• Schema validation workflow: Add structured data validation to your CI/CD pipeline or deployment process. Catch schema errors before they go live.

Each SOP should be a simple, step-by-step document that anyone on your team can follow. Include screenshots where helpful, link to the specific tools needed, and define clear success criteria. A content writer should be able to pick up your "New Content Workflow" document and produce an AI-optimized page without needing to ask questions. That is the test of a good SOP.

Also create a one-page AI search reference card with the fixed panel, source requirements, applicable schema checks, crawler-role distinctions, and pre-publish checklist. Keep it in the team workspace and assign an owner for updates.

Here is a sample pre-publish checklist you can adapt for your organization:

• Page has exactly one H1 tag that clearly describes the topic
• Headings describe the questions or decisions the section resolves
• The opening paragraph under each heading provides a direct, bounded answer
• Specific data points disclose their source, date, denominator, and material limitations
• Eligible JSON-LD matches the visible page and current provider documentation
• FAQPage schema is used only when the page and current eligibility rules support it
• FAQ sections contain real unanswered customer questions when they improve the page
• All paragraphs pass the "extraction test": they make sense in isolation
• Internal links connect this page to at least 3 related pages
• Meta description is unique, accurate, and under 160 characters
• Schema validates without errors in Google's Rich Results Test
• dateModified reflects the date of the latest material page change

Week 12: Train Your Team and Set Up Alerts

The final week of your 90-day plan focuses on two critical activities: knowledge transfer and early warning systems. Both are about making sure your GEO gains survive beyond the initial optimization sprint.

If you have a team, run a 30-minute training session covering:

• Why AI search visibility matters and how it differs from traditional SEO
• The content writing guidelines you developed (answer-first structure, question headings, self-contained paragraphs)
• How to add and validate JSON-LD schema markup
• How to read and interpret your AI Visibility dashboards
• The escalation process for when AI engines say something inaccurate about your brand

Set up automated alerts for critical changes:

• Visibility change alerts: Choose a threshold only after you know the panel's normal variance. An alert starts an investigation; it does not identify the cause.
• Competitor alerts: Monitor when new competitors start appearing in AI responses for your key queries.
• Inaccuracy alerts: Flag any AI response that contains incorrect information about your brand for immediate content correction.
• Schema error alerts: Monitor your structured data for validation errors using Google Search Console or automated testing tools.

Run one final comprehensive audit at the end of week 12. Compare your day-90 results to your day-1 baseline. Document everything: what worked, what did not, what you would do differently, and what your ongoing priorities are. This retrospective becomes the foundation for your monthly maintenance cycle.

Your week 12 audit should answer these questions:

• What is your final presence rate compared with day 1, using the same denominator?
• How many tested engines cite your brand for at least one fixed query?
• How many mentions are positive, neutral, or negative, with the denominator shown?
• Have you closed the gap with your top competitor? (Measure relative presence rate)
• Are your technical scores higher than day 1? (Run Foglift's Technical Audit again to compare)
• How many pieces of original research or data-driven content did you publish?
• How many external citations and backlinks did you earn?

Create a brief "90-Day AI Search Review" that shows the before-and-after panel, technical changes, published assets, independent-source events, referrals, and business outcomes as separate measurements. Include unchanged and negative results. The document becomes a reproducible record for the next cycle.

The work from these 12 weeks gives you a measured baseline, an evidence ledger, and a repeatable review loop. Whether visibility changes remains an empirical question for the next panel.

What to Keep After Day 90

Keep the fixed prompt-engine panel, technical checks, source ledger, and business outcomes on one repeatable review cadence. The detailed weekly work above remains the operating record, so there is no value in duplicating all 12 weeks as a second checklist.

• Re-run the same panel: Preserve prompts, engines, dates, and denominators so the next observation is comparable.
• Audit changed pages: Validate crawler access, rendered markup, and material facts after substantive releases.
• Work the source layer: Use cited-source evidence to choose one attainable independent corroboration opportunity.
• Review outcomes separately: Track mentions, citations, referrals, leads, and conversions without treating one metric as proof of another.

Use the AI search KPI guide for formulas and denominators, the AI search audit guide for the recurring technical review, and the AI search optimization checklist as the compact execution companion.

Five Principles That Make This Plan Work

These five principles keep the plan measurable and the published evidence reviewable:

1. Use a repeatable cadence. Keep the prompt panel, engine set, and review schedule stable enough to compare observation windows.
2. Measurement supports attribution. A fixed baseline and change ledger let you compare outcomes without inventing a multiplier.
3. Separate technical checks. Verify the relevant crawler policy, fetch result, indexing evidence, and rendered markup before attributing an answer gap to copy.
4. Original data needs a method. Publish the sample, field dates, calculation, limitations, and downloadable evidence. Then measure whether the artifact enters citation sets.
5. Keep entity facts consistent. Maintain one accurate source of truth for the brand, product, people, prices, and relationships, then inspect how answers describe them.

Clear visible content, accurate structured data, primary sources, and honest product facts support human review as well as machine extraction. Measure search, answer visibility, referrals, and conversions independently so one metric is not used as proof of another.

Common Mistakes to Avoid During Your 90 Days

These mistakes break the measurement loop or exceed what the cited provider documentation establishes:

• Skipping the baseline: Record the prompt, engine, target page, date, and denominator before choosing an intervention.
• Collapsing the engines: Foglift's frozen Q3 2026 panel found 0.094 mean pairwise cited-domain overlap across five engines. Report each engine separately.
• Treating access as an outcome: OpenAI documents OAI-SearchBot for search discovery and GPTBot for potential training use. The controls are independent. Verify the crawler that serves the tested feature; an allow rule does not prove crawling, indexing, or citation.
• Claiming movement from one answer: Keep the panel fixed and record at least 14 days before and after an intervention.
• Publishing weak evidence: Use real customer evidence, disclose relationships, and publish a method an editor or buyer can verify.
• Creating a duplicate page: Benchmark the existing canonical first, then repair a specific content, source, or markup gap.

Adapting This Plan to Your Situation

• Solo founder: Start with one product line, a small fixed panel, and the pages tied to those prompts.
• Large site: Extend the technical and content phases, assign owners, and sample pages by template and business value.
• E-commerce: Keep Product and Offer facts current, then test buyer prompts against the cited product sources.
• SaaS or B2B: Publish current plan boundaries and disclose the method behind every outcome number.
• Local business: Keep Business Profile facts accurate and use LocalBusiness markup only when it matches visible content.

The operating sequence stays useful across business types: establish the baseline, verify technical access, improve the intended page, inspect the source layer, measure the fixed panel, and document the next decision. Adapt the page set, prompts, and evidence standard to the business.

Getting Started Today

Start with Foglift's free Technical Audit and save the result. It records technical and AI Readiness findings for the scanned page. Pair it with a fixed prompt-engine panel for the day-one answer baseline.

Record each prompt, engine, answer, citation, date, and account context. The AI search optimization checklist is the compact execution companion. Day one produces a baseline; it does not promise that visibility will grow.

Sources & Further Reading

• Google Search Central, generative AI optimization guide. Direct source for technical structure, useful content, structured-data limits, and measurement.
• Google Search Central, AI features and your website. Direct source for Googlebot, index and snippet eligibility, visible content, and no special AI schema.
• OpenAI crawler documentation. Direct source for OAI-SearchBot, GPTBot, and ChatGPT-User roles.
• Anthropic crawler documentation. Direct source for Claude-SearchBot, ClaudeBot, and Claude-User controls.
• Perplexity crawler documentation. Direct source for PerplexityBot and Perplexity-User.
• Foglift's Q3 2026 citation benchmark. Frozen 375-answer panel, methodology, downloadable data, and 0.094 mean pairwise domain overlap.

Frequently Asked Questions

How long does it take to see results from AI search optimization?

There is no universal timeline. Crawl access, indexing, retrieval, and answer composition change independently across engines. Record a fixed prompt-engine panel for 14 days before a change, hold the panel constant, and compare at least 14 days afterward. Report each engine separately and preserve zero-result cells. The 90-day plan gives you several controlled observation windows; it does not guarantee a visibility increase.

Can I do AI search optimization myself or do I need an agency?

A solo founder or marketing team can run the measurement and content portions of this roadmap. Robots.txt, CDN, WAF, rendering, and schema changes may require a developer when the site stack or permissions make them technical. Foglift records Technical Audits and AI Visibility evidence; the site owner still decides and implements each change.

What is the most important thing to do in the first week?

The single most important first-week action is establishing your baseline. Run a Technical Audit, then run an AI Visibility Check across the engines your buyers use and document where your brand appears, what sources are cited, and where you are absent. Inspect robots.txt, CDN, and WAF rules for the search crawlers you intend to allow. An allow rule removes one possible access barrier; it does not prove crawling, indexing, or citation.

What should I do after the 90-day plan ends?

After completing the 90-day plan, keep the operating loop outlined above. Recheck technical access, refresh material facts, monitor the same priority prompts, and compare changes by engine. Keep a change log so later movement can be tied to a specific intervention without turning correlation into proof. A second 90-day sprint can target another product line, geographic market, or content vertical.

How much does AI search optimization cost?

The workflow can be run manually with a spreadsheet and free product access. Foglift provides unlimited single-page Technical Audits and active-use weekly Perplexity monitoring on Free. Launch starts at $49 per month and adds all five tracked engines, daily monitoring, and developer access. Team time depends on the number of prompts, pages, and source corrections in scope.

Which AI engine should I prioritize first?

Prioritize the engines your buyers actually use, then report each engine separately. Retrieval and source behavior differ across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overview, so a blended average can hide movement in one engine and a decline in another. No provider publishes a universal one-to-two-week reflection timeline for content changes.