AI SEO: What It Is, How It Works, and What to Measure

Originally reported 2026-09-03

Learn AI SEO: A Free 30-Day Evidence-First Practice Plan

Why it matters: Learn AI SEO through seven practical modules and a 30-day portfolio: foundations, crawler controls, claims, citations, referrals, experiments and automation.

Next check: Each concept becomes a small auditable artifact before use.

You can learn AI SEO by mastering four loops: learn a documented concept, test it on a controlled sample, verify the evidence, and apply only what survives review. The fastest path is not watching more tool demos; it is producing small, auditable work artifacts.

Use Google’s learning paths to check your foundations

Google’s October 6 learning-path announcement groups existing Search Central videos into beginner Search fundamentals and Technical SEO playlists. The first covers crawling, indexing, serving and measurement with Search Console and Trends; the second emphasizes directives, crawl analysis, rendering, duplicates and API or export workflows. It is a curated learning route, not a new ranking requirement.

For this guide’s practice plan, pair each lesson with one observation on your own site. After learning indexing, inspect one URL and distinguish recorded status from a live fetch. After the measurement lessons, save one date-filtered query/page view. After rendering or duplicate-content lessons, compare the user-facing page and canonical record. Keep the result even when it is “no change needed.”

Use the technical path when the blocker is infrastructure. Use the beginner path when the team is still mixing crawling, indexing and serving. Completing a playlist is not evidence of improved rankings. The value is being able to explain a concrete observation and choose the next diagnostic step. Google’s UGC pipeline and ChatGPT’s interactive answers are newer surface changes to evaluate only after those foundations are clear.

Module 1: Learn the search foundations

Study crawling, rendering, indexing, canonicals, snippets, internal links and measurement. Google’s SEO Starter Guide is a primary foundation. Then read its AI optimization guidance, noting which advice is documented and which popular claims are absent.

Exercise: audit ten URLs for status, canonical, index eligibility and clear primary answer. Deliver a one-page finding list with evidence.

Module 2: Learn platform controls

Compare OAI-SearchBot and GPTBot, PerplexityBot and Perplexity-User, and your CDN’s purpose-based controls. Record role, vendor source, robots token, verification method and intended policy. Avoid copying a universal robots.txt template.

Exercise: build a crawler control matrix and verify one authentic log request or mark it unknown. Use our AI crawler guide.

Module 3: Learn claim-level content

Take one existing page and classify sentences as documented fact, first-party observation, interpretation, recommendation or unknown. Replace secondary reporting with primary sources where available. Add one worked example and one meaningful limitation. Do not increase length unless the reader gains decision support.

Exercise: complete the workflow in our AI-search page rewrite guide and keep a before-and-after diff.

Module 4: Learn citation observation

Create 20 prompts from real user jobs. Freeze wording, locale and platform state for a cycle. Save answers, cited URLs and claim-support verdicts. Preserve failures and report numerator plus denominator. Repeat after a defined interval, not until the outcome looks good.

Exercise: produce a citation ledger with at least five manually verified source-claim relationships.

Module 5: Learn referral and outcome measurement

Configure known AI referral rules, campaign parameters, landing-page analysis and conversions. Document dark-traffic limits. Compare observed citations with detected visits without assuming every citation should generate a click or every direct visit came from AI.

Exercise: build a small report separating mention, citation, referral and revenue using our four-ledger framework.

Module 6: Learn controlled experimentation

Select comparable pages, define one treatment, capture a baseline and choose a review window. Prefer treatments that help readers even if no AI effect appears. Record simultaneous changes and avoid causal language when the design cannot isolate them.

Exercise: preregister a page-level test with hypothesis, metric, denominator, exclusion rule and rollback trigger.

Module 7: Learn responsible automation

Automate source collection, formatting and review handoffs before automating publication. Require source IDs, human approval, least-privilege credentials, idempotency and rollback. Test missing-source, duplicate-slug and failed-public-check paths.

Exercise: diagram a draft-only workflow and pass every failure case in staging. Our n8n SEO workflow provides the gates.

Your 30-day practice plan

1. Week 1: foundations and crawl/index audit.
2. Week 2: claim-led rewrite and source verification.
3. Week 3: frozen prompt study and citation validation.
4. Week 4: referral report, experiment design and retrospective.

Keep a portfolio containing the audit, evidence ledger, rewritten page, prompt sample, measurement report and test plan. That portfolio demonstrates judgment better than a certificate or tool badge.

Download the 30-day AI SEO practice plan. Replace the examples, attach proof for each exercise and mark a module complete only after the Verify step; not merely after reading it.

Keep the AI SEO guide and task hub as your syllabus index, then follow only the specialist route required by the evidence in your practice portfolio.

Return to the briefing overview

Originally reported 2026-09-03

Content Optimization for AI Search: A Page-Level Rewrite Workflow

Why it matters: Repair an existing page with a clear claim, primary evidence, worked example and limitation; then measure the result without erasing the baseline.

Next check: Measure a fixed window and preserve what the page already did well.

Content optimization for AI search means repairing a page so its main claim, evidence, example and limitation are clear; without padding it, flattening expert nuance or pretending a rewrite guarantees citation. The best candidate is an already useful page with a specific decision-support gap.

Select pages by opportunity and risk

Start with pages that already earn relevant impressions, referrals, citations, conversions or support usage. Add pages containing important facts that are outdated or hard to locate. Avoid rewriting everything at once; simultaneous sitewide changes destroy your baseline and can remove content readers already value.

Capture title, canonical, index state, query-page pairs, conversion data, citations and last-updated date before editing.

Write the page’s answer contract

In one sentence, state the question, audience, direct answer and boundary. Example: “For indexed pages, Google says no special AI schema is required for eligibility, but eligibility does not guarantee an AI-feature link.” This contract keeps the rewrite from drifting into a generic topic overview.

Compare the contract with your existing citation-ready content guidance and competing pages on the site.

Repair claims at sentence level

Classify important sentences as fact, first-party observation, interpretation, recommendation or unknown. Link time-sensitive product facts to primary documentation. Add date and scope. Remove citations that merely mention the topic but do not support the claim.

Google’s AI optimization guide recommends accurate, unique and useful content while warning against scaled low-value production. It does not require artificial chunks or a target word count.

Add the missing information gain

Information gain is not novelty theater. It can be a clearer synthesis, a verified correction or a reusable tool that saves the reader work.

Improve structure without writing for a robot

Lead with the answer. Use descriptive H2 headings, concise paragraphs, real tables and steps in the order they are performed. Keep definitions near first use. Add internal links only when the next page serves a different job. Delete repetitive summaries and FAQ questions already answered in the body.

Do not replace precise terminology with awkward synonyms to increase keyword variation.

Review the human and evidence risks

Have a subject-aware reviewer check whether the rewrite changed meaning, removed a caveat, exposed confidential data or overstated a result. Verify every external link and promised download. Compare the final draft with the prior version and keep a rollback copy.

Our AI content brief guide includes safeguards for assisted drafting.

Measure through a fixed post-edit window

Record the publication date and changed sections. Watch index status, relevant queries, engagement, conversions, observed citations and support feedback. Do not attribute movement to the rewrite when other major changes occurred. If the page loses useful performance, inspect the query and conversion mix before rolling back.

Download the AI-search page rewrite ledger. Replace the example rows and require an evidence ID, expected reader benefit, reviewer and rollback decision for every substantive change.

Use a claim-level before-and-after test

Consider a weak sentence: “AI search prefers authoritative, well-structured content.” It names no platform, evidence, decision, or limit. A useful revision might say: “Google says pages must be indexed and eligible for a snippet before they can appear as supporting links in its AI features; eligibility still does not guarantee selection.” The revision is not better because it is longer. It is better because the platform, requirement, outcome, and boundary can be checked.

Save the original sentence, revised sentence, supporting primary source, reader benefit, and review date in the ledger. Then inspect the paragraph around it: the source should support the exact claim, the page should explain what the reader does next, and nearby wording should not quietly turn eligibility into a promise. If the claim is based on your own observation, record the sample and conditions instead of presenting it as platform policy.

Repeat this method for the few claims that control the reader’s decision. A page with five precise, actionable claims can be more useful than one with fifty generic tips. For cluster planning and measurement boundaries, return to the AI SEO guide and task hub.

Return to the briefing overview

Originally reported 2026-09-03

Google AI Mode SEO: Map Query Fan-Out Without Publishing Thin Pages

Why it matters: Map related questions around one user problem, then decide which belong in a deep guide and which truly deserve a separate URL.

Next check: It helps map supporting questions but does not expose every hidden query.

Google AI Mode SEO is not a license to manufacture one page for every imagined follow-up question. Google says AI features may use query fan-out; issuing related searches across subtopics and data sources; but its guidance still favors useful, non-commodity pages and warns against scaled low-value content.

The practical move is to map related questions, group them by one user problem, and build the smallest set of pages that genuinely resolves those jobs.

What query fan-out changes

Google’s AI search optimization guide says AI features can use query fan-out to find supporting responses. This means a broad prompt may trigger searches for definitions, comparisons, constraints, recent changes and next steps. It does not mean publishers can see the full hidden query set or that each subquery deserves its own URL.

Treat fan-out as a research model: “What must the system and reader understand to answer this task responsibly?”

Build a fan-out map from observable evidence

Start with Search Console queries, site search, support logs, sales objections and repeated prompt patterns. For one core problem, list the likely subjobs: define, diagnose, compare, implement, validate and recover. Mark the source for every branch. Label invented branches as hypotheses until demand or user research supports them.

Our AI keyword research workflow shows how to combine search and grounding data without passing estimates off as measured demand.

Decide when a subtopic earns a separate page

A separate keyword is not enough. The new URL should earn its maintenance cost and provide information that would make the parent page unwieldy.

Write one useful guide, not a page factory

Lead with the direct answer, then cover prerequisites, method, example, validation and limits. Use a comparison table or worksheet where it saves the reader time. Link out to a narrower implementation page only when the reader needs a different tool or context.

Google’s spam policies prohibit scaled content abuse when many pages are produced primarily to manipulate rankings, regardless of whether automation or humans create them. Review our scaled-content risk guide before automating URL creation.

Test the cluster for cannibalization

Export queries and landing pages for the cluster. Look for multiple URLs swapping positions for the same intent, inconsistent canonicals, near-identical introductions and internal links with ambiguous anchors. Compare conversions too: two URLs can rank without serving the same decision.

When merging pages, preserve genuinely useful sections, redirect retired URLs and update internal links. Do not merge solely because a tool flags keyword overlap.

Measure outcomes fan-out cannot reveal

Google does not expose every hidden subquery used inside an AI response. Separate what is documented from what you infer. Monitor indexation, Search Console page/query trends, observed supporting links, referral behavior and conversions. Add annotations when templates, titles or product availability change.

Do not call an impression change an “AI Mode lift” unless the reporting surface and design support that attribution. Our AI impressions versus traffic guide explains the denominator problem.

A repeatable editorial decision

1. Name the user problem.
2. Collect observed subquestions and mark hypotheses.
3. Group by decision, evidence and workflow.
4. Choose one deep page or a small, linked set.
5. Add a testable artifact or example.
6. Review overlap quarterly and after major product changes.

Download the AI Mode query fan-out map. Replace its example rows, then use the “page decision” column to record why a branch belongs on the guide, deserves a separate URL or should not be published.

Return to the AI SEO guide and task hub to compare this Google AI Mode workflow with other answer surfaces and shared SEO foundations.

Return to the briefing overview

Originally reported 2026-09-03

Google AI Overviews SEO: What Makes a Page Eligible for Citations

Why it matters: Google documents ordinary Search eligibility for AI Overview supporting links; not a special schema or guaranteed inclusion. The workflow starts there.

Next check: Google documents these ordinary Search requirements for supporting links.

SEO for Google AI Overviews begins with ordinary Search eligibility: the page must be indexed and eligible to appear with a snippet. Google does not document a special AI schema, an AI text file or a guaranteed optimization recipe. The useful strategy is to remove eligibility blockers, answer a real query well, and measure supporting-link appearances separately from clicks.

Understand what Google actually documents

Google’s AI features and your website documentation says the standard technical requirements for Search apply to supporting links in AI Overviews and AI Mode. A page needs to be indexed and snippet-eligible. It also states that no special schema.org markup, machine-readable file or “AI text file” is required.

That establishes eligibility, not selection. Google explicitly notes that meeting requirements does not guarantee crawling, indexing, serving or appearance. Any provider promising a special tag that forces inclusion is selling beyond the documentation.

Remove the common eligibility blockers

• Confirm the preferred URL returns 200 and is not soft-404 content.
• Check robots.txt, noindex and X-Robots-Tag together.
• Verify the canonical points to the intended page and Google selected it.
• Make the main answer available in rendered HTML.
• Review nosnippet and max-snippet controls deliberately.
• Implement paywalled-content markup correctly where relevant.

Do this at template level, not one URL at a time. Our new-site SEO audit provides the broader technical sequence.

Build a page around the user’s decision

Google’s AI search optimization guidance recommends unique, non-commodity content and warns against scaled pages that add little value. Start with the decision a person must make. Give the direct answer, evidence, method, example and limit in a structure a reader can scan.

If the query asks whether a setting is required, state “required,” “optional,” or “not documented” near the top. If the answer varies, name the variable. Useful precision beats paragraphs filled with synonyms for the target keyword.

Design for supporting claims, not isolated chunks

A self-contained passage should remain part of a coherent page. Define the term, support factual claims with primary sources, add a worked example and connect the reader to the next action. You do not need to fragment a subject into dozens of tiny pages merely because retrieval systems analyze passages.

Use descriptive headings and tables where they reduce comparison effort. Avoid FAQ sections that repeat the article in question form. Our topic-cluster guide explains how to cover adjacent intent without cannibalization.

Measure AI Overview visibility cautiously

Save observations with query, country, device, account state, date, answer type, cited URL and screenshot or HTML evidence where permitted. Google interfaces change, personalization can matter and a single manual check is not a stable rank. Describe the sample and its limits.

In Search Console, analyze page and query performance, but do not infer that every impression or click came from an AI Overview unless the reporting surface explicitly provides that dimension. Use server and analytics evidence for downstream behavior.

Run a page-level improvement test

1. Select pages with meaningful demand and weak decision support.
2. Capture index, query and conversion baselines.
3. Add one substantive asset: original data, a decision table, test procedure or calculator.
4. Strengthen primary-source citations and label uncertainty.
5. Request recrawl only when appropriate, then observe through a fixed window.
6. Report changes without claiming causality the design cannot prove.

Avoid the five expensive myths

There is no documented AI Overview schema. There is no guaranteed word count. There is no evidence that adding “according to experts” creates authority. An AI crawler visit does not prove selection. And a third-party visibility score is not revenue. Each myth collapses a multi-stage system into an easy sales claim.

Download the AI Overviews eligibility and observation log. It keeps technical eligibility, observed source selection and business outcomes in separate columns so your next decision survives product changes.

Use the AI SEO guide and task hub to connect this Google-specific observation log with WordPress, content, reporting, and other platform workflows.

Return to the briefing overview

Originally reported 2026-09-03

ChatGPT SEO: How to Earn Discovery, Citations and Clicks

Why it matters: A practical ChatGPT SEO workflow that separates crawler discovery, answer citations, referral clicks and qualified outcomes instead of hiding them in one score.

Next check: A citation matters differently from a referral that produces a useful action.

ChatGPT SEO is the work of making a page discoverable by OpenAI’s search crawler, useful enough to be cited in an answer, and compelling enough to earn a qualified click. Those are three different outcomes. A crawl log can prove discovery; a cited answer can prove selection; analytics can show a referral. Treating them as one “AI visibility” score hides where the real problem sits.

This guide gives publishers a practical workflow for ChatGPT search without pretending that a special schema type, prompt trick, or crawler visit guarantees inclusion.

Start with the three-stage ChatGPT SEO funnel

OpenAI’s publisher FAQ documents OAI-SearchBot for search discovery and says ChatGPT referrals can carry utm_source=chatgpt.com. GPTBot is a separate control for model training. That distinction matters: allowing search discovery does not require making the same choice about training.

Make the page eligible before chasing citations

Confirm that the canonical URL returns a stable 200 response, is not blocked for OAI-SearchBot, presents the main answer in rendered HTML, and does not contradict itself across title, body and structured data. Then check the page in your normal search-indexing workflow. ChatGPT search is not a substitute for technical SEO hygiene.

Use the server log, not a user-agent screenshot alone. User-agent strings can be spoofed. Save timestamp, requested path, response code and, where possible, verified network identity. Our AI crawler verification guide explains the evidence chain.

Write citation-sized claims inside a useful page

A citation-worthy passage answers one precise question, states the scope, links to its evidence and names the limit. For example, “OpenAI documents OAI-SearchBot as the crawler used to surface websites in ChatGPT search” is testable. “OpenAI loves authoritative content” is vague and attributes an undocumented preference.

Do not split every possible subquestion into a thin URL. Build one page around a real job, then use descriptive headings, short definitions, examples and a decision table. The goal is not artificial “chunking”; it is making claims easy for humans and systems to interpret. See our guide to content answer engines can cite.

Test citations with a reproducible prompt set

Create prompts from customer questions, support tickets, Search Console queries and comparison tasks. Freeze the wording for a test cycle. Record platform, model or mode when visible, date, account state, locale, answer, cited URL, cited claim and whether the source actually supports that claim. Run the same prompts at planned intervals; do not refresh until you get the result you want.

Separate brand mention from citation. A response may name you without linking, cite you without placing the citation near the relevant claim, or link to a third party that mentions you. Those are different observations and should remain separate rows.

Measure clicks without erasing dark traffic

Build a GA4 exploration for explicit ChatGPT referrals and the documented campaign parameter, but keep a second view for landing-page conversions and self-reported discovery. Apps, privacy tools and copy-paste behavior can remove referrer data. Referral sessions are therefore a defensible lower-bound signal, not a complete count of influence.

Report sessions, engaged sessions, conversions and value by landing page. A citation that sends ten qualified trial users may matter more than a hundred answer appearances with no attributable action. Our GA4 AI referral workflow provides the analytics setup.

Use a 30-day ChatGPT SEO operating cycle

1. Week 1: resolve crawler, canonical, status and rendering issues.
2. Week 2: improve five high-intent pages with sourced claims, examples and explicit limits.
3. Week 3: run the frozen prompt set and validate every observed citation.
4. Week 4: reconcile citations with referrals and conversion outcomes; choose the next fixes.

Download the ChatGPT SEO discovery–citation–click ledger. Its sample rows are marked EXAMPLE-REMOVE; replace them before drawing conclusions. The useful outcome is not a flattering score. It is knowing whether the next constraint is access, answer usefulness, source selection, click appeal or conversion.

Compare this surface-specific process with the broader AI SEO guide and task hub before generalizing a ChatGPT observation to Google, Bing, or Perplexity.

Return to the briefing overview