Twenty short lessons with exercises. Work through them at your own pace; no account or paid service is required to read them.
A mention count measures visibility, while a lead or sale measures a business result. Choose one commercial outcome and one visibility measure before spending time on changes. A result can move without the other moving.
Try it: Write the outcome, its measurement source and a date for reviewing it.
Start with questions customers actually ask your team. Include the service and location where relevant. Separate a broad learning question from a comparison or booking question so a report explains which stage needs work.
Try it: List six questions and label each learn, compare or choose.
Models can confuse similarly named businesses. Publish a clear name, service area, website and contact details. Keep aliases documented so analysis can distinguish your name from an unrelated organization.
Try it: Check your homepage and one directory entry for inconsistent details.
Use businesses a customer could reasonably choose instead. A competitor share calculated from a selected set is not market share. Record why each rival belongs in the set and when you change it.
Try it: Choose three rivals and write one sentence explaining each choice.
AI answers vary between requests. Keep the exact prompts, engines, market and date of the initial measurement. Repeated samples are more informative than one screenshot, but they still have uncertainty.
Try it: Save the settings and several answers before editing your website.
A model API with web search can produce different answers from a consumer assistant. Providers also change models and search behavior. Label the source accurately and compare like with like across time.
Try it: Write the engine, access method and known limits alongside a sample answer.
A crawler cannot use information it cannot retrieve. Check page status, redirects and authentication requirements. A public page is different from a page that every bot will choose to index or cite.
Try it: Open an important page signed out and record its final URL and status.
Robots rules ask compliant crawlers to avoid paths; they do not stop people or hostile software. Specific user-agent groups can override wildcard groups. Protect private records with authentication.
Try it: Evaluate one private path for both a wildcard bot and a named crawler.
A focused service page should explain who the service helps, where it is available, the work included and important exclusions. Useful concrete answers help people evaluate an offer without inflated promises.
Try it: Rewrite one vague paragraph using the actual service and its boundaries.
Use headings that describe the question being answered. Put the direct answer near the heading and add examples or qualifications after it. Avoid hiding essential information inside images or inaccessible interactions.
Try it: Choose one customer question and add a concise factual answer to its page.
Structured data describes visible content in a machine-readable form. It should not invent reviews, prices, authors or availability. A generated document needs review and validation; no markup guarantees a rich result or citation.
Try it: Generate markup for one real page and check each field against what visitors see.
Names, dates, statistics and comparisons should have reliable sources. Keep the source close to a consequential claim, and remove a claim when its evidence no longer supports it. Original evidence is better than repeated assertions.
Try it: Find one numerical claim and save the source that supports its exact wording.
Hours, service areas, fees and staff details can become stale. An update date should mean a meaningful review occurred. Assign responsibility for important pages instead of adding automatic dates to unchanged text.
Try it: Choose five facts, an owner and a review interval for each.
A citation does not always recommend the cited brand. Read the surrounding answer and the destination page. Distinguish an owned page, a comparison, a directory and a critical mention before deciding what to improve.
Try it: Classify five citations and note whether each actually recommends the business.
A large checklist can obscure the next useful action. Start with a defect that affects customers and can be verified: an inaccessible page, misleading price, missing service detail or broken link.
Try it: Choose one fix and write its observable acceptance condition before changing it.
Generated content can invent facts or miss context. A human reviewer should check identity, services, prices, sources and legal commitments. Keep drafts separate from publishing permission and preserve a way to undo edits.
Try it: Mark every unsupported statement in a draft before considering publication.
Referral data can help identify assistant traffic, but missing referral headers and privacy controls make it incomplete. Avoid treating every unlabelled visit as AI traffic or every citation as a visit.
Try it: Compare one week of known referrals with the total traffic without extrapolating missing data.
Changing prompts, locations or engines changes what the score means. Compare only equivalent measurement scopes and annotate changes. Small shifts with few samples may be normal variation rather than an improvement.
Try it: Pick two reports and list settings that must match before comparing scores.
A useful experiment states a change, expected observable outcome and review window. Avoid several simultaneous changes if you want to understand their effects. Seasonal demand and competitors remain alternative explanations.
Try it: Write a one-change experiment with a measurement date and a reason it might fail.
A useful report separates observed data, estimates, assumptions and missing evidence. Show the customer what changed and what remains uncertain. Promise the work you control rather than a guaranteed ranking or earnings result.
Try it: Write a short report with one observation, one limit and one next action.