Track rankings, citations, and pipeline
Connect technical health, organic demand, AI visibility, source presence, and qualified pipeline in one operating view.
Build one scorecard that connects shipped work to visibility to pipeline, reporting share instead of absolutes and cohorts instead of screenshots, so the program survives its budget reviews.
You will build a scorecard that shows movement, diagnoses causes, and tells the team what to do next.
Rankings alone miss AI visibility; AI mentions alone miss commercial impact. A useful scorecard combines leading signals, visibility outcomes, and business results without pretending every touch can be perfectly attributed.
IF YOU SKIP IT / Programs do not usually die of bad execution; they die of unprovable results. A screenshot folder is not a trend line, and a trend line is what the board funds.
- Leading vs lagging
- Leading metrics prove the machine is running (work shipped, visibility moving). Lagging metrics prove it mattered (pipeline, revenue). They move on different clocks.
- Source register
- One row per data file: its scope, the exact filter, the tab, the row count, the figures it produced, and its caveats. It is what lets you re-derive a number months after you first quoted it.
- Unattributable
- Clicks or sessions that fall outside every filter you defined. Name them and report them on their own line rather than folding them into whichever side of the split looks better.
- Cohort
- A fixed group of pages or prompts you measure together over time, so improvements are not drowned in sitewide averages.
- Sourced vs assisted
- Sourced pipeline names your channel as the first touch; assisted means the channel appeared somewhere in the journey. Reported separately, always.
- Self-reported attribution
- Asking buyers directly how they heard of you (a form field, a sales question). Currently the cleanest signal AI influence leaves.
Each phase feeds the next. Don't skip ahead, the artifact at the end is only trustworthy if every phase ran.
- 01Define
Agree on metric name, formula, source, grain, window, owner, and decision before building a dashboard.
- 02Baseline
Freeze priority page, query, prompt, source, conversion, and pipeline cohorts.
- 03Instrument
Validate analytics events, CRM capture, consent behavior, and campaign metadata.
- 04Join
Connect work shipped to affected assets and demand families.
- 05Review weekly
Use leading indicators to catch implementation and discovery problems.
- 06Review monthly
Evaluate visibility, citations, actions, opportunities, and confounders.
- 07Decide
Record continue, expand, repair, test, or stop decisions with the evidence used.
- 08Explain
Report uncertainty, lag, model volatility, and attribution limits directly.
- ✓You now have
Search visibility scorecard
Check what your data actually covers
Most wrong conclusions in this work are correct arithmetic on data that covers something other than what you thought.
Every wrong number in search reporting looks right. The arithmetic checks out, the chart renders, nobody catches it, because the mistake happened before the maths: the file covered something other than what you thought. On one diagnosis, five separate Search Console exports all carried the same page filter, address contains the product path. That is not one page. It was eight, and nothing in the filenames said so.
So spend the first twenty minutes not analyzing. Open the Filters tab on every export and write what it actually covers at the top of the file. Then count the rows, because Search Console stops at exactly 1,000, which means every share you calculate from it is a share of the top 1,000 and needs saying out loud each time you quote one.
- 01
For every export, write the exact filter, the tab it came from, the row count, and the date range at the top of the file before you analyze anything.
- 02
Confirm which figures are page-level and which are folder-level, then label every number you quote with the one it is.
- 03
Check that your branded and unbranded filters actually complement each other, and report the clicks that fall in neither as unattributable.
- In Search Console, open the Filters tab inside the export. A page filter reading "contains /product/" covers every URL under that path, and only the Pages tab separates them. Chart, Queries, Countries, and Devices are all path-wide.
- Count the rows. Exactly 1,000 means the export is truncated, so note it beside every percentage you derive from that file.
- Test that the branded and unbranded filters complement each other. Filtering branded on "acme" and unbranded on "-acm" leaves a gap. Run both totals against the unfiltered one and report the difference as unattributable rather than folding it into whichever side looks better.
Every export carries its filter, tab, row count, and date range at the top, and for any number in your scorecard you can say whether it describes one page or a whole folder.
- WATCH OUT
Quoting a folder figure as a page figure is undetectable in the output. It inflates every number and survives every later check, because the arithmetic really is correct.
- PRO TIP
Keep a source register: one row per file with its scope, the figures it produced, and its caveats. When someone challenges a number in month three, you re-derive it in a minute instead of an afternoon.
OPTIONAL / AI CO-PILOTRun this lesson with AI, the prompt, sources and expected output+
Choose the business priority, interpret exceptions, protect confidential data, challenge weak evidence, and approve the final decision.
Clean, classify, compare, calculate, and draft rows from supplied evidence. AI may surface patterns; it does not own strategy.
GSC + rank tracker, plus aeo visibility from Citation tracker.
| File | Exact filter | Tab used | Rows | Truncated | Figures it produced | Caveat |
|---|---|---|---|---|---|---|
| gsc-product-queries.csv | URL contains /product/ | Queries | 1,000 | Yes | Non-brand click share | Folder-level, 8 pages |
CLAUDE / CHATGPT PROMPTAnalyze the evidence without outsourcing the decision+
Remove personal or confidential data. Attach the named exports, explain every column, and tell the model when the dataset was collected.
You are assisting a human B2B SaaS SEO and AEO operator with Module 09: Track rankings, citations, and pipeline. Current lesson: Check what your data actually covers Objective: Most wrong conclusions in this work are correct arithmetic on data that covers something other than what you thought. Required artifact: Search visibility scorecard BUSINESS CONTEXT I WILL PROVIDE - Product, category, target market, pricing model, sales motion, and geography - The priority customer segment and the commercial outcome for this 90-day cycle - Demand visibility exported from GSC + rank tracker - AEO visibility exported from Citation tracker - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. For every export, write the exact filter, the tab it came from, the row count, and the date range at the top of the file before you analyze anything. 2. Confirm which figures are page-level and which are folder-level, then label every number you quote with the one it is. 3. Check that your branded and unbranded filters actually complement each other, and report the clicks that fall in neither as unattributable. REQUIRED OUTPUT Return a table using these exact columns: File | Exact filter | Tab used | Rows | Truncated | Figures it produced | Caveat. For every recommendation, cite the source row, URL, call note, or data point that supports it. Add a confidence column in your analysis: High, Medium, or Low. List missing evidence separately instead of guessing. Finish with a section named HUMAN DECISIONS REQUIRED. RULES - Do not invent search volume, revenue, customer statements, product capabilities, or competitor facts. - Do not treat correlation as causation. - Preserve contradictory evidence and explain why it conflicts. - Do not make the final priority or publishing decision. Prepare the evidence for a human owner. - Use this example only as a format reference, not as evidence: A Search Console export filtered to “URL contains /product/” covers every page under that path, not the single page the question was about.
Every figure names the file, tab, and filter it came from. A named human owner must verify this before the lesson is complete.
A Search Console export filtered to “URL contains /product/” covers every page under that path, not the single page the question was about.
Separate leading and outcome metrics
Shipping and visibility move at different speeds.
A farmer who dug up seeds every week to check on them would harvest nothing, yet that is how most AEO programs get judged: planted in January, dug up in February, cancelled in March. Work, visibility, and revenue grow on three different clocks. You ship this week, visibility moves next month, pipeline moves next quarter.
The scorecard has to respect the clocks, so it gets three labeled layers: what we shipped, what the market saw, what the business got. One field note while you build it: AI referral traffic will look embarrassingly small for years. Read its conversion rate instead of its volume. It is routinely the highest-intent segment you have, hiding behind the smallest number on the report.
- 01
Track leading work: fixes shipped, pages improved, passages added, source wins, and reviews earned.
- 02
Track visibility: index coverage, ranking distribution, non-brand clicks, answer inclusion, citation share, and accuracy.
- 03
Track business outcomes: qualified CTA actions, assisted opportunities, sales usage, and influenced pipeline.
- Build the scorecard as one sheet with three labeled sections: execution (from your board), visibility (from Search Console and the baseline sheet), business (from the CRM).
- In GA4, create the AI referral segment: session source contains chatgpt.com, perplexity.ai, gemini.google.com, or copilot.microsoft.com. Watch its conversion rate, not its session count.
- Give every metric three cells: formula, source, owner. A metric missing any of the three gets deleted.
Your scorecard has three labeled layers, and nobody in the review confuses a shipped fix with a business result, or dismisses AI referrals by volume.
- PRO TIP
Present the three layers in order every time: what we shipped, what the market saw, what the business got. The order teaches stakeholders the lag.
- WATCH OUT
Reporting layer-three numbers weekly trains everyone to panic on noise. Business metrics move monthly at best.
OPTIONAL / AI CO-PILOTRun this lesson with AI, the prompt, sources and expected output+
Choose the business priority, interpret exceptions, protect confidential data, challenge weak evidence, and approve the final decision.
Clean, classify, compare, calculate, and draft rows from supplied evidence. AI may surface patterns; it does not own strategy.
Citation tracker, plus behavior cohorts from GA4 / product analytics.
| Metric | Definition | Formula | Source | Grain | Window | Owner |
|---|---|---|---|---|---|---|
| Tier-1 citation rate | Runs with owned citation | cited runs / valid runs | Citation tracker | Prompt-engine | 28 days | AEO lead |
CLAUDE / CHATGPT PROMPTAnalyze the evidence without outsourcing the decision+
Remove personal or confidential data. Attach the named exports, explain every column, and tell the model when the dataset was collected.
You are assisting a human B2B SaaS SEO and AEO operator with Module 09: Track rankings, citations, and pipeline. Current lesson: Separate leading and outcome metrics Objective: Shipping and visibility move at different speeds. Required artifact: Search visibility scorecard BUSINESS CONTEXT I WILL PROVIDE - Product, category, target market, pricing model, sales motion, and geography - The priority customer segment and the commercial outcome for this 90-day cycle - AEO visibility exported from Citation tracker - Behavior cohorts exported from GA4 / product analytics - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Track leading work: fixes shipped, pages improved, passages added, source wins, and reviews earned. 2. Track visibility: index coverage, ranking distribution, non-brand clicks, answer inclusion, citation share, and accuracy. 3. Track business outcomes: qualified CTA actions, assisted opportunities, sales usage, and influenced pipeline. REQUIRED OUTPUT Return a table using these exact columns: Metric | Definition | Formula | Source | Grain | Window | Owner. For every recommendation, cite the source row, URL, call note, or data point that supports it. Add a confidence column in your analysis: High, Medium, or Low. List missing evidence separately instead of guessing. Finish with a section named HUMAN DECISIONS REQUIRED. RULES - Do not invent search volume, revenue, customer statements, product capabilities, or competitor facts. - Do not treat correlation as causation. - Preserve contradictory evidence and explain why it conflicts. - Do not make the final priority or publishing decision. Prepare the evidence for a human owner. - Use this example only as a format reference, not as evidence: A new comparison page may show indexation and source pickup before it produces a closed opportunity.
Leading, visibility, and business metrics are separated. A named human owner must verify this before the lesson is complete.
A new comparison page may show indexation and source pickup before it produces a closed opportunity.
Keep crawled, cited, referred, and converted apart
Four states, four proofs, and none of them implies the next one.
A page can rank first and earn nothing. That is not a figure of speech. On one diagnosis a page ranked first for searches worth 1,500 a month and sent zero visits, while the flagship product page appeared 109,265 times for 13 clicks. Position moved in both cases. Nothing followed it.
That keeps happening because four different things get reported as one. Crawled means a bot fetched the page. Cited means an engine named you in an answer. Referred means somebody actually arrived. Converted means a real sales conversation happened. Each has its own proof, and each has a long list of things it does not prove, so give them four columns and never let them collapse into a single visibility number.
A verified request in the server log: bot named, response code recorded, date kept.
Nothing about being read. A page can be fetched daily for a year and never appear in an answer.
Named in an answer, against a stated prompt, engine, and date, with the cited URL saved.
Nothing about a visit. The reader may have got what they needed and never clicked.
A session attributed to an assistant, landing on a named page.
Nothing about value, and it undercounts, because assistants strip referrers.
A qualified sales conversation traced back to that session.
Usually unmeasurable on day one. The definition is a request, not a metric.
A page can rank first and earn nothing. On one diagnosis, a page ranked first for searches worth 1,500 a month and sent zero visits, and the flagship product page appeared 109,265 times for 13 clicks. Position moved in both. Nothing behind it moved at all.
- 01
Record crawled from server logs: the bot named, the response code, and the date.
- 02
Record cited against a stated prompt, engine, and date, and referred as a session landing on a named page.
- 03
Record converted against a written definition of what the event counts, and keep all four in separate columns permanently.
- Crawled: take it from server logs or your CDN, not from a tool estimate. Record the bot name, the response code, and the date. A 200 for Googlebot says nothing about OAI-SearchBot.
- Cited: record the prompt, engine, date, and cited URL from the Module 06 capture sheet. Referred: use the GA4 AI referral segment defined in the previous lesson. Neither one implies the other.
- Converted: before building anything, ask what each recorded conversion event counts, which of them represents a real sales conversation, and whether it can be split by landing page and by branded against non-branded.
Crawled, cited, referred, and converted sit in four separate columns, each with the evidence behind it, and you have a written definition of what a conversion event counts.
- PRO TIP
Ask for the conversion definition in week one. It is the cheapest request in the whole program, it changes what every number below it means, and being met with silence is itself the finding.
- WATCH OUT
Referral attribution undercounts by design, because assistants strip referrers. A small number in that column is not evidence that AI sends nobody.
OPTIONAL / AI CO-PILOTRun this lesson with AI, the prompt, sources and expected output+
Choose the business priority, interpret exceptions, protect confidential data, challenge weak evidence, and approve the final decision.
Clean, classify, compare, calculate, and draft rows from supplied evidence. AI may surface patterns; it does not own strategy.
GA4 / product analytics, plus pipeline cohorts from HubSpot / Salesforce.
| Page or prompt | Crawled | Cited | Referred | Converted | Evidence | What it does not prove |
|---|---|---|---|---|---|---|
| /agency-reporting/ | 200, OAI-SearchBot, 2026-08-14 | 3 of 12 runs | 4 sessions | Undefined | Log + capture sheet + GA4 | Citation says nothing about a visit |
CLAUDE / CHATGPT PROMPTAnalyze the evidence without outsourcing the decision+
Remove personal or confidential data. Attach the named exports, explain every column, and tell the model when the dataset was collected.
You are assisting a human B2B SaaS SEO and AEO operator with Module 09: Track rankings, citations, and pipeline. Current lesson: Keep crawled, cited, referred, and converted apart Objective: Four states, four proofs, and none of them implies the next one. Required artifact: Search visibility scorecard BUSINESS CONTEXT I WILL PROVIDE - Product, category, target market, pricing model, sales motion, and geography - The priority customer segment and the commercial outcome for this 90-day cycle - Behavior cohorts exported from GA4 / product analytics - Pipeline cohorts exported from HubSpot / Salesforce - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Record crawled from server logs: the bot named, the response code, and the date. 2. Record cited against a stated prompt, engine, and date, and referred as a session landing on a named page. 3. Record converted against a written definition of what the event counts, and keep all four in separate columns permanently. REQUIRED OUTPUT Return a table using these exact columns: Page or prompt | Crawled | Cited | Referred | Converted | Evidence | What it does not prove. For every recommendation, cite the source row, URL, call note, or data point that supports it. Add a confidence column in your analysis: High, Medium, or Low. List missing evidence separately instead of guessing. Finish with a section named HUMAN DECISIONS REQUIRED. RULES - Do not invent search volume, revenue, customer statements, product capabilities, or competitor facts. - Do not treat correlation as causation. - Preserve contradictory evidence and explain why it conflicts. - Do not make the final priority or publishing decision. Prepare the evidence for a human owner. - Use this example only as a format reference, not as evidence: A page can be fetched daily and never cited, then cited for months and never referred.
Crawled, cited, referred, and converted never merge into one number. A named human owner must verify this before the lesson is complete.
A page can be fetched daily and never cited, then cited for months and never referred.
Build stable cohorts
Aggregate sitewide numbers can hide what changed.
We watched a live domain sit in third place across all 690 tracked keywords, and first place, at nearly double the nearest competitor, across the 205-keyword cohort the program actually worked on. Same site, same week, same tool. The sitewide average was hiding the entire result.
That is what cohorts are for. Group pages by type and improvement date, group prompts by tier, and compare each cohort only to itself over time. Then annotate the timeline with every release and campaign, because in month three nobody remembers what shipped when, and an unexplained spike gets credited to luck instead of to you.
- 01
Group pages by type, topic, segment, and publish/improvement date.
- 02
Group prompts by tier, archetype, segment, and engine.
- 03
Compare the same cohorts over time and annotate releases, campaigns, and major search changes.
- In Search Console, isolate cohorts with page filters or regex (Performance > Pages > filter). In rank trackers like Semrush, tag keyword groups and read the tag-level visibility.
- Name cohorts by date and intent ('2026-Q3 comparison pages') and freeze membership on day one.
- Keep an annotations tab: every release, campaign, and known algorithm update with its date, so spikes have explanations when you read the chart in month three.
Improved pages and tracked prompts live in named, frozen cohorts, and the timeline carries annotations for every release that could explain a move.
- PRO TIP
The blended-vs-cohort chart in this lesson is worth recreating with your own data once. It converts skeptics in one slide.
- WATCH OUT
Adding new pages into a cohort mid-quarter breaks the comparison. New pages start a new cohort.
OPTIONAL / AI CO-PILOTRun this lesson with AI, the prompt, sources and expected output+
Choose the business priority, interpret exceptions, protect confidential data, challenge weak evidence, and approve the final decision.
Clean, classify, compare, calculate, and draft rows from supplied evidence. AI may surface patterns; it does not own strategy.
HubSpot / Salesforce, plus demand visibility from GSC + rank tracker.
| Cohort | Start date | Pages | Queries / prompts | Baseline | Target | Notes |
|---|---|---|---|---|---|---|
| Q3 money pages | 2026-07-01 | 8 | 34 / 24 | 22% visibility | 40% | Exclude brand terms |
CLAUDE / CHATGPT PROMPTAnalyze the evidence without outsourcing the decision+
Remove personal or confidential data. Attach the named exports, explain every column, and tell the model when the dataset was collected.
You are assisting a human B2B SaaS SEO and AEO operator with Module 09: Track rankings, citations, and pipeline. Current lesson: Build stable cohorts Objective: Aggregate sitewide numbers can hide what changed. Required artifact: Search visibility scorecard BUSINESS CONTEXT I WILL PROVIDE - Product, category, target market, pricing model, sales motion, and geography - The priority customer segment and the commercial outcome for this 90-day cycle - Pipeline cohorts exported from HubSpot / Salesforce - Demand visibility exported from GSC + rank tracker - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Group pages by type, topic, segment, and publish/improvement date. 2. Group prompts by tier, archetype, segment, and engine. 3. Compare the same cohorts over time and annotate releases, campaigns, and major search changes. REQUIRED OUTPUT Return a table using these exact columns: Cohort | Start date | Pages | Queries / prompts | Baseline | Target | Notes. For every recommendation, cite the source row, URL, call note, or data point that supports it. Add a confidence column in your analysis: High, Medium, or Low. List missing evidence separately instead of guessing. Finish with a section named HUMAN DECISIONS REQUIRED. RULES - Do not invent search volume, revenue, customer statements, product capabilities, or competitor facts. - Do not treat correlation as causation. - Preserve contradictory evidence and explain why it conflicts. - Do not make the final priority or publishing decision. Prepare the evidence for a human owner. - Use this example only as a format reference, not as evidence: Track the eight improved comparison pages separately from 300 historical blog URLs.
Pages and prompts use stable cohorts. A named human owner must verify this before the lesson is complete.
Track the eight improved comparison pages separately from 300 historical blog URLs.
Connect search to pipeline honestly
B2B journeys are multi-touch and dark.
Here is a real buyer journey: heard the founder on a podcast, asked ChatGPT for a shortlist, clicked your comparison page, came back four days later through brand search, and told the form 'a colleague recommended you'. Four signals, all true, none complete. B2B attribution is dark and multi-touch, and pretending one number captures it destroys trust in every number you present.
So keep all the signals. Report sourced and assisted pipeline as separate lines, add ChatGPT and Perplexity to your 'how did you hear about us' dropdown, and once a month trace five real opportunities end to end. Self-reported attribution is unfashionable, and it is also the cleanest trace AI influence leaves today.
Sourced and assisted pipeline never share a cell. Add ChatGPT, Perplexity, and AI search to the “how did you hear about us” dropdown, and trace five real opportunities end to end every month.
- 01
Capture landing page, self-reported source, known content touches, and sales-confirmed influence.
- 02
Report sourced and assisted pipeline separately.
- 03
Review relevant account journeys qualitatively each month to find repeatable paths.
- Add the dropdown to your demo and signup forms: how did you hear about us, with ChatGPT, Perplexity, AI search, Google, podcast, colleague as options.
- Ask sales to log any mention of AI assistants in call notes with a consistent tag, and pull the tag count monthly.
- Once a month, trace five recent opportunities end to end (first touch, self-reported source, content touched) and write three sentences about the pattern.
Sourced and assisted pipeline never share a cell, the form field lists AI assistants as options, and monthly journey reviews are on the calendar.
- PRO TIP
Keep the form field optional and short. A five-option dropdown gets answered; a twelve-option one gets skipped.
- WATCH OUT
When self-reported and tracked attribution disagree, keep both. The disagreement is information about dark touchpoints, not an error to resolve.
OPTIONAL / AI CO-PILOTRun this lesson with AI, the prompt, sources and expected output+
Choose the business priority, interpret exceptions, protect confidential data, challenge weak evidence, and approve the final decision.
Clean, classify, compare, calculate, and draft rows from supplied evidence. AI may surface patterns; it does not own strategy.
GSC + rank tracker, plus aeo visibility from Citation tracker.
| Layer | Metric | Baseline | Current | Delta | Threshold | Decision |
|---|---|---|---|---|---|---|
| Visibility | Top-10 demand families | 22% | 36% | +14 pts | >=35% | Continue |
CLAUDE / CHATGPT PROMPTAnalyze the evidence without outsourcing the decision+
Remove personal or confidential data. Attach the named exports, explain every column, and tell the model when the dataset was collected.
You are assisting a human B2B SaaS SEO and AEO operator with Module 09: Track rankings, citations, and pipeline. Current lesson: Connect search to pipeline honestly Objective: B2B journeys are multi-touch and dark. Required artifact: Search visibility scorecard BUSINESS CONTEXT I WILL PROVIDE - Product, category, target market, pricing model, sales motion, and geography - The priority customer segment and the commercial outcome for this 90-day cycle - Demand visibility exported from GSC + rank tracker - AEO visibility exported from Citation tracker - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Capture landing page, self-reported source, known content touches, and sales-confirmed influence. 2. Report sourced and assisted pipeline separately. 3. Review relevant account journeys qualitatively each month to find repeatable paths. REQUIRED OUTPUT Return a table using these exact columns: Layer | Metric | Baseline | Current | Delta | Threshold | Decision. For every recommendation, cite the source row, URL, call note, or data point that supports it. Add a confidence column in your analysis: High, Medium, or Low. List missing evidence separately instead of guessing. Finish with a section named HUMAN DECISIONS REQUIRED. RULES - Do not invent search volume, revenue, customer statements, product capabilities, or competitor facts. - Do not treat correlation as causation. - Preserve contradictory evidence and explain why it conflicts. - Do not make the final priority or publishing decision. Prepare the evidence for a human owner. - Use this example only as a format reference, not as evidence: An opportunity reports “podcast,” first visited a comparison page, and later returned through brand search; preserve all three signals.
Sourced and assisted pipeline are not conflated. A named human owner must verify this before the lesson is complete.
An opportunity reports “podcast,” first visited a comparison page, and later returned through brand search; preserve all three signals.
Write the monthly decision narrative
A dashboard without interpretation does not operate the program.
A dashboard shows numbers; it cannot argue. The monthly narrative is where the program defends itself: what moved, why you believe it moved, how confident you are, what did not move, three next actions, and one thing you will stop doing.
Write it in sentences, because sentences force the honesty charts let you skip, and state your confidence out loud. 'Medium confidence, could also be the pricing change' builds more credibility than certainty theater. The stop-doing line matters most. A program that only ever adds workstreams is six months from collapsing under its own roadmap.
- 01
State what moved, why you believe it moved, confidence level, and what did not move.
- 02
Show the evidence chain from shipped work to visibility to buyer behavior.
- 03
Choose three next actions and one thing to stop or defer.
- Use a fixed monthly template: what moved, why we believe it moved, confidence (high/medium/low), what did not move, three next actions, one stop.
- Paste the two most important charts inline and write the narrative in sentences, not bullets. Thirty minutes, sent to the same list every month.
- Keep every monthly narrative in one running doc; by month six it is the program's story, and the renewal case writes itself.
This month's narrative names its evidence chain and its confidence level, and it retired at least one activity that was not earning its keep.
- PRO TIP
Write the confidence level honestly. 'Medium confidence, could also be the pricing change' builds more trust than certainty theater.
- WATCH OUT
The fifteen-chart appendix is where decisions go to die. Two charts and a paragraph beat a dashboard tour.
OPTIONAL / AI CO-PILOTRun this lesson with AI, the prompt, sources and expected output+
Choose the business priority, interpret exceptions, protect confidential data, challenge weak evidence, and approve the final decision.
Clean, classify, compare, calculate, and draft rows from supplied evidence. AI may surface patterns; it does not own strategy.
Citation tracker, plus behavior cohorts from GA4 / product analytics.
| Review date | Evidence | Confounder | Decision | Owner | Next check | Confidence |
|---|---|---|---|---|---|---|
| 2026-08-07 | Visibility +14 pts | Brand campaign | Expand cluster | Growth lead | 2026-09-04 | Medium |
CLAUDE / CHATGPT PROMPTAnalyze the evidence without outsourcing the decision+
Remove personal or confidential data. Attach the named exports, explain every column, and tell the model when the dataset was collected.
You are assisting a human B2B SaaS SEO and AEO operator with Module 09: Track rankings, citations, and pipeline. Current lesson: Write the monthly decision narrative Objective: A dashboard without interpretation does not operate the program. Required artifact: Search visibility scorecard BUSINESS CONTEXT I WILL PROVIDE - Product, category, target market, pricing model, sales motion, and geography - The priority customer segment and the commercial outcome for this 90-day cycle - AEO visibility exported from Citation tracker - Behavior cohorts exported from GA4 / product analytics - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. State what moved, why you believe it moved, confidence level, and what did not move. 2. Show the evidence chain from shipped work to visibility to buyer behavior. 3. Choose three next actions and one thing to stop or defer. REQUIRED OUTPUT Return a table using these exact columns: Review date | Evidence | Confounder | Decision | Owner | Next check | Confidence. For every recommendation, cite the source row, URL, call note, or data point that supports it. Add a confidence column in your analysis: High, Medium, or Low. List missing evidence separately instead of guessing. Finish with a section named HUMAN DECISIONS REQUIRED. RULES - Do not invent search volume, revenue, customer statements, product capabilities, or competitor facts. - Do not treat correlation as causation. - Preserve contradictory evidence and explain why it conflicts. - Do not make the final priority or publishing decision. Prepare the evidence for a human owner. - Use this example only as a format reference, not as evidence: “Citation share rose after two source wins” is a hypothesis; name the matching prompts and URLs, then retest.
The report ends with decisions and owners. A named human owner must verify this before the lesson is complete.
“Citation share rose after two source wins” is a hypothesis; name the matching prompts and URLs, then retest.
The workspace shows a working scorecard: the metric dictionary, a baseline cohort, the weekly view, and, the part most teams never build, the decision log. Every metric has a formula, a source, and an owner. If a number on your scorecard cannot fill those three fields, it is decoration.
SEO + AEO evidence stack
Connect shipped work to rankings, citations, buyer actions, and pipeline without claiming causality the data cannot support.
Leading-to-lagging indicator movement
Read the sequence: work shipped, discoverability, visibility, buyer action, then pipeline.
Check each item only after the artifact meets the standard. Progress saves on this device; use the academy-home backup to move it elsewhere.