The Moat Map course

Module 1: The reader and their task

Who is this site for, and what are they trying to get done?

Every site I run starts here. Your queries show what people asked and hint at the task behind it. They can't tell you who the person is, so write that part down as a guess. Pick the reader whose task your site serves best and your evidence backs.

Suds & Snouts, a dog groomer: 60% of clicks come from people booking soon, so its reader is someone booking this week. Pinecone Payroll picks people switching payroll mid-year, though its tax calculator gets more clicks.

Open Search Console: Performance, last 3 months, Queries. Ask AI to group your searches by the task behind them, then check any ambiguous rows.

Check your work: A stranger can tell your evidence from your guesses.

Analyze the evidence: The reader and their task

Attach your query export with its dates and filters, plus the pages people land on. If you have no report, use Harbor’s initial input instead.

Prompt for your AI tool

Help me with: Who is this site for, and what are they trying to get done?

Group the searches by the decision a person is trying to make. Keep source rows and totals. Suggest a primary reader/task and a second plausible task; label audience guesses. Explain which existing page serves each. Give me one source row and one calculation to verify, then stop for my choice. After I choose, draft a reader card using the labels on this sheet.

Download the fictional practice input

Harbor practice: Name one person and their decision, and mark what you're guessing.

Your sheet

Your reader card

• Who they are (a guess) Label it as a guess. Fictional example: A homeowner whose furnace stopped
• Their task Fictional example: Get a technician out today
• A real search, in quotes Add the report and date range. Fictional example: "furnace repair ashford", last 3 months
• Where they arrive from Fictional example: Google, mostly on a phone
• What the first screen must answer Fictional example: Can someone come today, and what's the number?
• What the report can't show Fictional example: Who they are, or whether they called
• Evidence Report, date range, and the date the data runs through. Fictional example: Top 25 searches, Aug to Oct, data through Oct 31

Second reader

• Who, and their task Fictional example: A homeowner comparing heat pump rebates
• How their first screen differs Fictional example: Rebate amount and deadline come first

Module 2: Own or concede

Which searches can your site own, and which should you leave to someone else?

Ask AI to group searches by the task and map the existing answers. Then choose: Own means invest in an answer you can support; Concede means leave it to a better existing answer; Watch means get more evidence. A minimum search count can limit workload, but it does not prove a content gap or make a small, important task worthless.

Cobalt Comfort tests a rebate eligibility check built from the official rules, with a link to the authority. It leaves basic heat-pump definitions to manufacturers.

Ask AI to find the patterns first. Check ambiguous groups, then choose which searches to own, concede or watch.

Check your work: Every Own row says what today's best answer misses. Every Concede row names who already does it well.

Analyze the evidence: Own or concede

Attach your query export, current page extracts and the reader card you chose in module 1. Include dates and filters; add examples of the competing answers you actually inspected.

Prompt for your AI tool

Help me with: Which searches can your site own, and which should you leave to someone else?

Group the searches and summarize their evidence first. Suggest a demand floor with its limits as one input to my priority decision; do not stop parsing to seek approval. Propose Own, Concede or Watch for each group using current coverage and supplied competing answers, not counts alone. Show totals, row references, today’s best answer and what is missing. Mark uninspected answers unknown. Give me one calculation and one coverage claim to check; stop for my decision.

Harbor practice: The venue owns the facts. Decide what's left for Harbor to own.

Your sheet

Your demand floor

Set it before you sort.

• Times shown Fictional example: 150
• Within days Fictional example: 90
• Date you set it

Your searches

For each search, record:

• Search, in quotes Fictional example: "furnace repair ashford"
• Impressions Fictional example: 3000
• Clicks Fictional example: 150
• Who answers it best today Fictional example: Directories
• Your call Options: Own, Concede, Watch.
• Why, in one line Fictional example: We can come today; a directory can't

Module 3: The moat ladder

How much work would a competitor need to copy this page, and what move makes it more useful?

Start with what the person cannot decide on the current page. Ask AI for two different ways to remove that work, including a simple fix. Compare usefulness, evidence, effort and upkeep. Choose an answer you can build and test. A new format alone is not a lasting advantage.

Summit & Sparrow adds fit notes from 2K of its own boot returns, which moves a template category up a rung. The Dice Pantry leaves its hours page alone.

Start with your top five pages by clicks, or the pages behind your Own searches.

Check your work: Each move gives the reader a better answer, using data, access or an angle you already have.

Analyze the evidence: The moat ladder

Attach your chosen searches, page copy or screenshots, and a list of data, access or skills your brand actually has. Include your available hours.

Prompt for your AI tool

Help me with: How much work would a competitor need to copy this page, and what move makes it more useful?

Start with the reader decision. Identify what the current answer forces that person to work out. Generate two distinct ways to remove that work, including a plain correction or source link. For each, sketch the first useful answer, identify the evidence/capability needed, estimate effort and explain what a competitor would still need to reproduce after seeing it. Do not treat novelty as a moat. Give me the evidence claim to check, then stop for my choice. After I choose, draft one page change and a box sketch.

Harbor practice: Place the directory listing, the venue notice and your idea on the ladder.

Your sheet

Your top pages on the ladder

For each page, record:

• Page Fictional example: The 40 city pages
• Reader (module 1) Fictional example: Homeowner, furnace stopped
• Rung today Options: Template, Assembled data, Creative reframing.
• What a competitor needs to copy it Fictional example: A day: same paragraph, city name swapped
• One move up a rung Fictional example: Add local permit rules and rebates
• Hours Fictional example: 12

The page you'll move up this month

• Page Fictional example: The 10 busiest city pages
• Rung today Options: Template, Assembled data, Creative reframing.
• Rung next Options: Template, Assembled data, Creative reframing.
• The one move Fictional example: Per-city facts: permits, rebates, install time

Module 4: A promise you keep

What will your site keep current, and which original source supports it?

Check each claim against its original source: your job records for work you did, or the issuing authority for a permit rule. Comparing two copies of the same report is not a second check. Update the check date after reviewing that source.

Halcyon Valley Health checked doctor availability against its own export. Every check passed while 1 in 5 doctors weren't taking patients. Now it checks the scheduling system.

Write it so someone else could run the check without asking you.

Check your work: Someone else could run this check, and it fails on a wrong fact.

Analyze the evidence: A promise you keep

Attach the page claims you plan to publish, their authoritative source extracts and your available people/tools. Include source dates.

Prompt for your AI tool

Help me with: What will your site keep current, and which original source supports it?

Map each important claim to its supplied original or authoritative source, including the brand’s own records when that is where the fact comes from. Propose a realistic check owner, cadence, visible checked date, correction note and fallback if the source changes. Draft one known-wrong value to test the check; distinguish a proposed test from one actually run. Give me a source match to verify. Stop for my decision before writing the commitment card.

Harbor practice: Name who checks the venue's notices, and what the page does if one changes.

Your sheet

Your commitment card

• What you keep current, and how often Fictional example: Rebate amounts and deadlines, on the 1st of each month
• Checked against Name the original record or authority. Fictional example: Tri-County Power's rebate page
• Who checks, and how Fictional example: Marketing coordinator, from a calendar reminder
• Where "last checked" shows Fictional example: Under each amount, with the date
• Where corrections show Fictional example: A short list at the bottom of the page
• If the source changes or disappears Fictional example: Take the amounts down and say "call us to confirm"

Test the check

For each fact, record:

• One fact on the page Fictional example: Rebate: $1,000
• What the check compares it to Fictional example: The utility's rebate page
• A wrong value to test with Fictional example: Last year's $1,500
• When you tested it Options: Caught it, Missed it.

Module 5: Said from your own work

What can your site say from its own work, and does every feature work with one reader?

Useful original work can give someone a reason to visit beyond a summary. Show the decision your data, tested idea or experience helps them make, and test that usefulness rather than assume it. Prefer features that work for one reader; an empty public count can make a useful page look abandoned.

Ghost Hollow Hot Sauce cut star ratings that sat at three or fewer per sauce. It added a heat chart its staff retaste every batch.

Pick one line you'll stand behind, then test your features against an empty room.

Check your work: Every line could turn out wrong, and you've written down when you'll check it.

Analyze the evidence: Said from your own work

Attach your chosen reader task, page draft and evidence you gathered yourself. Describe any planned interactive or community features.

Prompt for your AI tool

Help me with: What can your site say from its own work, and does every feature work with one reader?

Suggest one useful contribution grounded in the supplied work. Explain the specific decision it helps, what a short summary can already answer and what useful work remains on the page. Uniqueness alone is not a reason to click. Compare each feature with a simpler version that works for one reader. Identify unsupported claims and one task-test idea. Ask me to verify a source claim and choose what to keep before drafting the feature.

Harbor practice: Write what Harbor can say from its own work.

Your sheet

The line you'll stand behind

• The line From your own data, access or history. Fictional example: The average age of furnaces we replaced, per county
• Date you'll grade it

Said from your own work

For each line, record:

• The line or feature Fictional example: Furnace age per county
• Where it comes from Fictional example: Last year's job records
• Where it lives Fictional example: Each county's city pages
• Would people still click past a summary? Why? Fictional example: Compare local furnace ages to decide whether to repair or replace; test whether that helps the choice

The empty-room test

For each feature, record:

• A feature you have or plan Fictional example: Ask-a-tech forum
• Works with one reader? Options: Yes, No.
• How it looks on day one, empty Fictional example: No posts, looks abandoned
• Keep, change or cut Options: Keep, Change, Cut.

Module 6: Get found

Which pages that Google skips deserve work, and which are fine as they are?

Export the Page indexing report. Most rows are fine, like redirects you meant. Look for your module 2 and 3 pages that Google crawled but didn't keep, and real pages returning 404 that people still link to. Redirect only to a real replacement.

Quillfox moved its help center and redirected 260 old articles to their new versions. The 40 about retired features stay 404.

Work from the report you exported. Give every row a bucket.

Check your work: Every Fix row is a page from your own sheets, and every other row has a reason to leave it.

Analyze the evidence: Get found

Attach the Page indexing export, relevant URL Inspection/live-check results if available, and the pages you chose to improve. Keep dates beside each report.

Prompt for your AI tool

Help me with: Which pages that Google skips deserve work, and which are fine as they are?

Parse URLs by purpose, reported status and observed live behavior. Suggest Intended, Fix, Investigate or Leave with row references; distinguish stale report data from current checks. Prioritize the chosen reader pages within my time limit. Draft redirects only where a relevant replacement is supplied. Do not claim you fetched a URL or ran a test. Give me one source row and one proposed action to check, then stop for my release decision.

Your sheet

Sort the report

For each url, record:

• URL Fictional example: /coupons/spring-tuneup/
• Status in the report Fictional example: Not found (404)
• Moat page? Options: Yes, No.
• Bucket Options: Intended, Fix, Investigate, Leave.
• Next action Fictional example: Redirect to the rebate page
• Check date

Redirects for pages that moved

For each redirect, record:

• Old URL Fictional example: /coupons/spring-tuneup/
• New URL Or "none, leave as 404". Fictional example: /rebates/
• Why it's a fair replacement Fictional example: Same offer, current amount
• Tested

Module 7: The first phone screen

Does the right page get the search, and can a phone reader see why it's worth choosing?

Open each moat page on a phone. The first screen should show the answer, a reason to trust it and the next step. Then check which page Google shows for each search, and whether it's the one you meant.

Pinecone Payroll's pricing page hid the price under a table three screens long. Now a team-size picker with the price sits on the first screen.

Grab your phone. Check each page before you scroll.

Check your work: A stranger holding your phone finds the answer and the next step without scrolling.

Analyze the evidence: The first phone screen

Attach a 390px first-screen screenshot, page text and query/page export for your chosen page. Add your reader card and the proposed page change.

Prompt for your AI tool

Help me with: Does the right page get the search, and can a phone reader see why it's worth choosing?

Identify the answer, reason to trust and next action actually visible on the supplied first screen. Map each search to its observed landing page. Compare a plain edit with one alternative layout; draft actual first-screen words and a labeled box sketch. Do not claim a screenshot proves reader understanding. Give me one screenshot/source claim to check; stop for my choice. Propose one uncoached task and a visible failure to watch for.

Harbor practice: Write the first words a visitor sees, and sketch the panel in boxes.

Your sheet

First-screen check at 390px

For each page, record:

• Page Fictional example: City page
• The reader's question Fictional example: Can someone come today?
• Answer on screen 1? Options: Yes, No.
• Reason to trust on screen 1? Options: Yes, No.
• Next step on screen 1? Options: Yes, No.
• What pushes it down Fictional example: A hero photo and a long intro
• Fix Fictional example: Call button above the photo
• Retest

Which page does Google pick?

For each search, record:

• Search, in quotes Fictional example: "furnace repair ashford"
• Page Google shows most Search Console, Pages tab. Fictional example: /blog/furnace-tips/
• Page you meant Fictional example: /ashford/
• Your call Options: Keep, Merge, Update, Leave.

Module 8: Measured, and the log

Did this change follow the calendar or the site, and what do you write down?

On a seasonal site I test the calendar first. Compare complete windows of equal length, and the same weeks last year if you have them. Write your call in the log before you know the result, with the date you'll check.

Gridiron Gremlin's clicks fell 80% after the championship game, and 75% in the same weeks last year. That pattern supports checking seasonality; it does not prove the cause or show that a rewrite is needed.

Check one change against the calendar, then write this month's note.

Check your work: Every claim in your note says what you saw. Nothing says "caused" without a test behind it.

Analyze the evidence: Measured, and the log

Attach the before/after report rows, dates/filters, change log and relevant calendar or prior-year data if you have it. Keep sessions, clicks and citations separate.

Prompt for your AI tool

Help me with: Did this change follow the calendar or the site, and what do you write down?

Parse and compare complete equal windows; show row references, denominators and calculations. Identify calendar changes, site changes and measurement changes without claiming causation. Offer the strongest competing explanations and the missing evidence that separates them. Give me one calculation and one source fact to verify. Stop for my judgment, then draft a short decision log entry, next action and check date.

Harbor practice: Compare each search with its own two equal windows.

Your sheet

Calendar or site?

For each change, record:

• The change you saw Metric, and both windows. Fictional example: City page clicks down 30%, compared with the previous four weeks
• Equal, complete windows? Options: Yes, No.
• What the calendar did Fictional example: Same weeks last year: down 5%
• What the site did Fictional example: Merged 12 city pages on Sept 15
• Your verdict Options: Calendar, Site, Can't tell.
• How you'll test it, and when Fictional example: Check the 12 redirects this week

Your monthly note

Five lines, same day every month.

• What changed With the date range and latest date included. Fictional example: Clicks down 30% on city pages, Sept 15 to Oct 12
• What I think it means Label it as your read. Fictional example: Working explanation: the merge may have broken some redirects
• What I did about it Fictional example: Checked all 12 redirects, fixed 3
• What I'm not claiming Fictional example: That the fix caused any recovery
• Next check date

Module 9: Your map

Where does your site compete, what's your one move this month, and how will you know in 28 days?

Use your saved sheets to choose one move. Ask AI for the first-screen copy and a box sketch, check and edit it, then keep the reader task and next check. Harbor is optional; reuse completed checks.

Tick each item as you do it.

Check your work: A stranger reads your map and can say who it's for, where it competes and what you'll check on day 28.

Analyze the evidence: Your map

Use your saved sheets and map as context. Attach the source evidence and current page screenshot for the one move you chose. Reuse completed checks; do not rerun every lesson.

Prompt for your AI tool

Help me with: Where does your site compete, what's your one move this month, and how will you know in 28 days?

Find contradictions or missing support in my map. Propose a single move with a smaller alternative and a realistic 30-day plan. Give me one source claim and calculation to verify, then wait for my decision. After I choose, draft the actual first-screen copy and a labeled box sketch for my own page, plus a task someone can try and the source/check date. Keep a planned test separate from observed results. List what I can ship now and what still needs evidence.

Harbor practice: Use your Moat Map as the page brief it asks for.

Ready to try your draft? Keep the original, revise a second version and record what a reader can do. Download the free page workbook. On your computer, unzip it and open the HTML file in your browser. It works offline with text copies; your drafts stay on that device. Download a backup from the workbook before closing it.

Your sheet

Your own page: the first draft

Use AI after your evidence check. Edit the draft yourself; a plan alone is not the finished artifact.

• First-screen copy Fictional example: The actual answer, reason to trust it and next useful action
• Box sketch Fictional example: Answer to source or uncertainty to next action. Describe the boxes or paste a file location.
• Reader task and next check Fictional example: What someone should find or decide; failure to watch for; when you will check. Mark it untested until observed.

Finish your map

• Check that boxes 1 to 5 filled from your sheets
• Reuse your saved indexing check, or name the evidence still missing
• Reuse your first-screen and page/query checks; do not repeat completed work
• Reuse your measured comparison and keep its limits
• Write your 30-day plan on the map
• Keep the decision this plan depends on, with a check-in date

Proof checks, in order

These fill box 6 of your map.

• Found Is the page indexed? Status and date. Fictional example: 36 city pages crawled, not indexed
• Understood Right page for the search? Answer on the first phone screen? Fictional example: Call button moved above the photo
• Chosen Shown and clicked? Date range. Fictional example: 2% of impressions become clicks
• Measured Calendar or site? AI answers citing it? Fictional example: Loss sits on the 12 merged pages

Your page plan

Keep the reader, answer, supporting evidence and next check together. Bring your module sheets into one document with your first-screen copy, box sketch, 30-day plan and check date.

Decision log

Record your recommendation, the source behind it and the date to check what happened. Keep observations separate from your interpretation.

The finished map

Each module's sheet