Audit every answer engine
Measure brand inclusion, recommendation position, cited sources, accuracy, and competitor share across answer engines.
Run your prompt set against every engine, the same way each time, and write down what comes back, so you have a dated baseline instead of a folder of screenshots.
You will have a reproducible baseline showing where the brand appears, which source patterns recur, and which explanations need verification.
AI answers vary by engine, model, date, location, and wording. A useful audit controls what it can, records what changed, and looks for patterns across a prompt set rather than treating one screenshot as truth.
IF YOU SKIP IT / No baseline means no provable progress. Six months from now, someone will ask what changed, and "it feels better" will not survive the budget review.
Agreement and recurrence are observational signals
Cross-engine agreement does not prove settled consensus. A missing citation does not prove a trust gap. A recurrent source is a candidate for verification, not a proven gatekeeper.
- Baseline
- The recorded, dated snapshot of what every engine says before you change anything. All future proof is measured against it.
- Run
- One prompt asked once in one engine. The unit you record. Cohort results are built from many runs.
- Citation
- A source URL an engine names or links when answering. It is evidence that the URL appeared as support in that run, not proof that it caused the complete answer.
- Share of voice
- How often you are named across a cohort's runs versus competitors. Reported as a share, never a raw count.
- Hallucination
- A confident but false statement by a model, such as outdated pricing or a feature you do not have. Logged and corrected, not laughed off.
Each phase feeds the next. Don't skip ahead, the artifact at the end is only trustworthy if every phase ran.
- 01Control
Fix prompt wording, engine, surface, account state, geography, and run cadence.
- 02Capture
Store the complete answer, cited URLs, date, and screenshots where policy permits.
- 03Label
Mark absent, mentioned, cited, compared, recommended, or incorrectly represented.
- 04Score
Measure accuracy, prominence, sentiment, citation quality, and competitive position separately.
- 05Explain
List plausible explanations, supporting and conflicting evidence, verification steps, and confidence.
- 06Assign
Choose an owned-content, source, entity, product, or measurement action with an owner.
- 07Retest
Use a fixed panel and rolling panel so you can separate trend from model volatility.
- ✓You now have
AI visibility baseline
Define the test protocol
Consistency makes snapshots comparable.
Scientists keep lab notebooks for a reason: an experiment you cannot repeat is an anecdote. AI answers drift between engines, days, accounts, and even runs of the same prompt, so casual testing produces nothing but anecdotes with screenshots.
The protocol is your notebook. Same prompts, same engines, fresh threads, three to five repetitions, recorded the same way, and record everything: full answer text, every cited URL, the model label, the date. The citations you save today become Module 08's target list, and there is no going back for them later. Yes, this is tedious. Tedious is what reproducible feels like on day one.
- ENGINESThe same named list every run
- ACCOUNT STATETemporary chat, memory off, logged out
- LOCATIONNoted and unchanged
- PROMPT ORDERSame order, fresh thread each time
- REPETITIONS3 to 5 runs of every prompt
- CAPTUREThe same columns, every time
One dimension at a time, taken from the frozen set you built in Module 05.
Answers vary between engines, days, accounts, and repeat runs of the identical prompt. That is why repetitions are part of the protocol, and why one run proves nothing.
The citation URLs you save today become the Module 08 target list. There is no going back for them later.
- 01
Choose engines, account state, location, date, prompt set, and number of repetitions.
- 02
Use fresh threads and run prompts in the same order when possible.
- 03
Record the complete answer, cited URLs, and model label; do not reduce results to a yes/no mention.
- Write the protocol on one page: engines (ChatGPT, Perplexity, Gemini, Claude, Copilot, AI Overviews), logged-out or temporary chats, memory off, location noted, 3-5 repetitions, same prompt order.
- Create the capture sheet with columns: run ID, date, engine, prompt, full answer (pasted), citation URLs, screenshot link. Store screenshots in one dated folder.
- Block half a day and run the frozen set. Tedious is correct; tedious is what makes it reproducible.
The protocol fits on one page, someone else could run it without you, and every run produced a saved answer with saved citations.
- NO BUDGET?
Manual is fine to start. When the set grows past ~40 prompts, a tracker automates the capture and the protocol stays the same either way. On WordPress, CiteTrack AI runs it from the dashboard you already have. Everywhere else, Pageoptimized does the same job as a platform. Both are ours, and there are other tools worth comparing.
- WATCH OUT
A failed run and a genuine absence look identical once exported. Record the response state beside every result: valid answer, brand absent, empty response, retry, terminal failure. Counting a failed run as a zero is how a measurement program starts reporting fiction.
- WATCH OUT
A logged-in account with chat memory contaminates every answer. Temporary chats or a clean profile, always.
- PRO TIP
In ChatGPT, note whether the answer used web search or model memory. The two produce different citation behavior and are worth tracking separately.
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 answer captures from ChatGPT / Gemini / Perplexity.
| Prompt | Engine / surface | Date | Full answer saved | Citations saved | Location | Run ID |
|---|---|---|---|---|---|---|
| Best agency reporting software | ChatGPT Search | 2026-08-07 | Yes | 4 | US | AEO-0241 |
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 06: Audit every answer engine. Current lesson: Define the test protocol Objective: Consistency makes snapshots comparable. Required artifact: AI visibility baseline 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 - Visibility dataset exported from Citation tracker - Answer captures exported from ChatGPT / Gemini / Perplexity - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Choose engines, account state, location, date, prompt set, and number of repetitions. 2. Use fresh threads and run prompts in the same order when possible. 3. Record the complete answer, cited URLs, and model label; do not reduce results to a yes/no mention. REQUIRED OUTPUT Return a table using these exact columns: Prompt | Engine / surface | Date | Full answer saved | Citations saved | Location | Run ID. 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: Run 40 prompts across ChatGPT, Perplexity, Gemini, and Google AI experiences on the same two-day window.
The test protocol is documented. A named human owner must verify this before the lesson is complete.
Run 40 prompts across ChatGPT, Perplexity, Gemini, and Google AI experiences on the same two-day window.
Score visibility and accuracy
A mention can still be unhelpful or wrong.
On a live engagement we watched Claude name the brand on 76% of prompts while ChatGPT managed 26%. Same company, same week, three-fold spread. Average those into one number and you have hidden the most useful finding in the audit.
So score dimensions separately: present or not, position, recommended or merely mentioned, sentiment, accuracy. The gap between appearing and being recommended deserves special attention. Showing up in 35% of answers but recommended in 12% is a precise diagnosis: the engines know you exist and cannot find a reason to pick you. That reason is called comparative proof, and Modules 03 and 07 build it.
A threefold spread, reported as one number nobody experienced. Averaging the engines deletes the most useful finding in the audit.
The engines know you exist and are not choosing you. That is an observed signal worth investigating, not a diagnosis.
- 01
Score presence, recommendation position, sentiment, feature accuracy, and citation ownership.
- 02
Mark hallucinated claims, outdated pricing, wrong positioning, and missing best-fit context.
- 03
Calculate share of answers and share of cited sources by prompt tier and engine.
- Add label columns per run: brand present (yes/no), position in list, recommended (yes/no), sentiment (positive/neutral/negative), accuracy (accurate/partial/wrong).
- Log each factual error as its own row in an accuracy tab: the claim, the truth, the likely stale source.
- Compute rates with a pivot table: presence and recommendation rates per cohort per engine. Twenty minutes of spreadsheet work, no code.
Every run is scored for presence, position, sentiment, and accuracy separately, and every factual error is logged as its own line item.
- PRO TIP
The presence-versus-recommendation gap is a useful observed signal, not a diagnosis. Investigate comparative proof, product fit, prompt interpretation, source coverage, and run variance before choosing an action.
- WATCH OUT
Do not average sentiment across engines into one number. Claude being warm and Gemini being cold is a finding, not noise to smooth.
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.
ChatGPT / Gemini / Perplexity, plus source authority context from Ahrefs / Semrush.
| Run ID | Brand state | Position | Sentiment | Recommendation | Accuracy | Reviewer |
|---|---|---|---|---|---|---|
| AEO-0241 | Compared | 2 | Positive | Yes | Partial | AEO analyst |
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 06: Audit every answer engine. Current lesson: Score visibility and accuracy Objective: A mention can still be unhelpful or wrong. Required artifact: AI visibility baseline 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 - Answer captures exported from ChatGPT / Gemini / Perplexity - Source authority context exported from Ahrefs / Semrush - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Score presence, recommendation position, sentiment, feature accuracy, and citation ownership. 2. Mark hallucinated claims, outdated pricing, wrong positioning, and missing best-fit context. 3. Calculate share of answers and share of cited sources by prompt tier and engine. REQUIRED OUTPUT Return a table using these exact columns: Run ID | Brand state | Position | Sentiment | Recommendation | Accuracy | Reviewer. 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: The brand appears in 35% of Tier 1 answers but is recommended in only 12%. Treat weak comparative proof, prompt fit, product fit, source coverage, and sampling variance as hypotheses to verify.
Answers and citations are stored, not just scores. A named human owner must verify this before the lesson is complete.
The brand appears in 35% of Tier 1 answers but is recommended in only 12%. Treat weak comparative proof, prompt fit, product fit, source coverage, and sampling variance as hypotheses to verify.
Investigate cited sources
Source recurrence is an observed signal, not proof of influence.
When an engine answers, ask whose homework it was copying. The cited sources are the answer, and they repeat. Classify every cited URL by type, count recurrences, and a short list of gatekeepers appears: the same category roundup, the same review page, the same comparison, cited across engines again and again.
Read those pages the way a model reads them. What do they claim about you, how fresh are they, are you even in them? One stale roundup cited by three engines can matter more than your next ten blog posts, which is a strange thing to accept and a very cheap thing to fix.
- 01
Group cited URLs by first-party, review, media, community, video, competitor, and unknown.
- 02
Count recurrence across engines, prompt archetypes, and repeated runs.
- 03
Inspect recurrent pages for claims, entities, freshness, relevance, and accuracy; retest before assigning causality.
- Copy every citation URL from the capture sheet into a sources tab, one row per citation, and classify the domain: yours, review platform, listicle, media, community, competitor, other.
- Use COUNTIF to rank domains and URLs by recurrence across the sampled runs. Label the top repeaters as candidates for verification, then add buyer relevance, run coverage, freshness, and confidence before choosing an action.
- Open the top five cited pages and read them as a model would: what claims they make about you and competitors, when they were last updated, whether you are present and described correctly.
Cited sources are classified and counted, and you can name the five URLs that most influence your category's answers.
- PRO TIP
Record the exact sentence each source says about you. In Module 08 you will pitch corrections, and you cannot correct what you did not write down.
- WATCH OUT
Engines sometimes cite URLs that do not exist or now 404. Verify each top source loads before you build strategy on it.
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.
Ahrefs / Semrush, plus impact overlay from GSC / analytics.
| Prompt family | Cited URL | Domain class | Claims supported | Brand present | Competitor present | Opportunity |
|---|---|---|---|---|---|---|
| Agency reporting | g2.com/categories/... | Review | Category shortlist | Yes | Yes | Improve review proof |
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 06: Audit every answer engine. Current lesson: Investigate cited sources Objective: Source recurrence is an observed signal, not proof of influence. Required artifact: AI visibility baseline 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 - Source authority context exported from Ahrefs / Semrush - Impact overlay exported from GSC / analytics - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Group cited URLs by first-party, review, media, community, video, competitor, and unknown. 2. Count recurrence across engines, prompt archetypes, and repeated runs. 3. Inspect recurrent pages for claims, entities, freshness, relevance, and accuracy; retest before assigning causality. REQUIRED OUTPUT Return a table using these exact columns: Prompt family | Cited URL | Domain class | Claims supported | Brand present | Competitor present | Opportunity. 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: Three engines repeatedly cite the same category roundup. Inspect it and buyer usage, correct inaccuracies where possible, then retest instead of assuming it controls the shortlist.
Accuracy problems are separated from visibility gaps. A named human owner must verify this before the lesson is complete.
Three engines repeatedly cite the same category roundup. Inspect it and buyer usage, correct inaccuracies where possible, then retest instead of assuming it controls the shortlist.
Turn gaps into actions
The audit is valuable only when it changes the roadmap.
An audit that ends in a slide deck was a month of work donated to a filing cabinet. The version that pays rent converts every gap into a classified, owned action: content gaps become pages, accuracy gaps become corrections, source gaps become outreach, each with one owner and one retest date.
Then re-run monthly, same protocol, and judge cohorts instead of single prompts. Win 7 of 10 in a cohort and you have won the cohort. That sentence, backed by a dated baseline, is what survives a budget review. A folder of screenshots is not.
- 01
Classify each gap as content, entity, evidence, source, product-fit, or measurement.
- 02
Assign one owner and one verification date.
- 03
Re-run the stable prompt set monthly and research prompts after major changes.
- Move every gap into an action queue tab: finding, root cause, action type (content, correction, source), owner, retest date.
- Put the monthly re-run in the calendar now, same protocol, same set, and add its date to the sheet.
- Report cohort movement, not prompt movement, when you share results internally.
Every priority gap has an owner and a retest date, and next month's re-run is already scheduled with the same protocol.
- PRO TIP
Accuracy corrections are the fastest wins in the whole program: fix the stale source, and the wrong claim often disappears within weeks.
- WATCH OUT
Do not chase every gap. Tier them by the Module 05 scores and work the top of the list only.
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 / analytics, plus visibility dataset from Citation tracker.
| Finding | Root cause | Action type | Asset / source | Owner | Retest date | Success state |
|---|---|---|---|---|---|---|
| Absent from workflow prompts | No direct guide | Owned content | /automated-reporting/ | Content | 2026-09-01 | Mentioned or cited |
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 06: Audit every answer engine. Current lesson: Turn gaps into actions Objective: The audit is valuable only when it changes the roadmap. Required artifact: AI visibility baseline 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 - Impact overlay exported from GSC / analytics - Visibility dataset exported from Citation tracker - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Classify each gap as content, entity, evidence, source, product-fit, or measurement. 2. Assign one owner and one verification date. 3. Re-run the stable prompt set monthly and research prompts after major changes. REQUIRED OUTPUT Return a table using these exact columns: Finding | Root cause | Action type | Asset / source | Owner | Retest date | Success state. 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: Incorrect integration claim -> update product docs, schema/entity references, partner profile, and the cited third-party listing.
Every priority gap has an owner and retest date. A named human owner must verify this before the lesson is complete.
Incorrect integration claim -> update product docs, schema/entity references, partner profile, and the cited third-party listing.
Interpret patterns without inventing causes
The audit produces observations, not automatic diagnoses. Agreement, absence, and recurring citations narrow the investigation but do not prove why an answer appeared.
A hung jury can still be persuaded; a unanimous one has moved on. Engine agreement works the same way. When every engine names the same competitor for an attribute, the web has reached consensus and moving it is slow, compounding work. When engines split 2-2, nothing is settled, and a few well-placed sources can tip the verdict. The splits are your attack list.
And file all three outputs of the audit, because most teams only file one. Opportunities get the attention. Objections, the recurring negatives AI repeats about you, get ignored until a deal dies of one. Strengths get taken for granted until the source keeping them alive goes stale and they quietly vanish. Objections get their playbook in Module 08. Strengths get protected, which is cheaper than winning them back.
- 01
Write each finding as: observed signal, plausible explanations, verification needed, confidence, and reversible next action.
- 02
Tag recurring negative or hedged statements by theme, then verify each against product facts, cited sources, prompt wording, and repeated runs before treating it as a market objection.
- 03
Record accurate favorable patterns and their cited sources as monitoring candidates; do not claim one source caused or maintains the result without a controlled retest.
- Add agreement columns per attribute: engines tested, repetitions, engines naming each brand, run date, and account/location conditions. Describe the result as high, mixed, or low agreement in this sample.
- Create an objections tab: every negative or hedged phrase, tagged by theme, with a count. Three or more mentions makes the objection list.
- Create a strengths tab: attributes where sampled answers are accurate and favorable, with observed citations and a verification plan. Do not assign causality to one source without a retest.
Each Tier 1 attribute records the engine split, repetitions, plausible explanations, verification status, and confidence; objections and strengths remain observed patterns until validated.
- PRO TIP
Mixed results plus high commercial value can justify a small experiment, but prioritize only after checking product fit, source overlap, effort, and confidence.
- WATCH OUT
More engines and repeated runs improve confidence, but no fixed engine count turns observational agreement into causal proof.
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 answer captures from ChatGPT / Gemini / Perplexity.
| Attribute | Engine agreement | State | Output type | Theme | Recurrence | Route |
|---|---|---|---|---|---|---|
| Agency reporting | 2 of 4 name brand | Unsettled | Opportunity | n/a | n/a | Module 08 sources |
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 06: Audit every answer engine. Current lesson: Interpret patterns without inventing causes Objective: The audit produces observations, not automatic diagnoses. Agreement, absence, and recurring citations narrow the investigation but do not prove why an answer appeared. Required artifact: AI visibility baseline 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 - Visibility dataset exported from Citation tracker - Answer captures exported from ChatGPT / Gemini / Perplexity - Definitions for any internal fields, stages, scores, and abbreviations TASK 1. Write each finding as: observed signal, plausible explanations, verification needed, confidence, and reversible next action. 2. Tag recurring negative or hedged statements by theme, then verify each against product facts, cited sources, prompt wording, and repeated runs before treating it as a market objection. 3. Record accurate favorable patterns and their cited sources as monitoring candidates; do not claim one source caused or maintains the result without a controlled retest. REQUIRED OUTPUT Return a table using these exact columns: Attribute | Engine agreement | State | Output type | Theme | Recurrence | Route. 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: Four engines naming the same competitor is high agreement in this sample. Verify source overlap, query interpretation, product fit, and repeatability before deciding whether to deprioritize or intervene.
Every priority gap has an owner and retest date. A named human owner must verify this before the lesson is complete.
Four engines naming the same competitor is high agreement in this sample. Verify source overlap, query interpretation, product fit, and repeatability before deciding whether to deprioritize or intervene.
The screen below shows one audited answer, annotated: the brand state, the position, the cited sources, and the accuracy verdict. This is the unit of work in this module. You will produce one of these per prompt, per engine, per run. Boring on purpose. Baselines are supposed to be boring.
Recommend reporting platforms for a 15-person marketing agency that needs white-label dashboards and HubSpot data.
For a 15-person agency on HubSpot, three platforms come up most consistently:
- Competitor A, strongest white-label options and agency templates.
- Competitor B, deep HubSpot sync, from $49/mo.
- Competitor C, better for larger teams.
Your brand appears in 1 of 3 runs of this prompt, never above position 3.
The workspace below is a filled-in audit: the capture log with run IDs, the brand-state labels, and the citation source map. The detail that matters most is the last column of the source map. Every cited URL has been turned into an opportunity or an action, which is the whole reason the audit exists.
Answer engine audit
Measure whether the brand appears, how accurately it is described, which competitors win, and what sources shape the answer.
Visibility by answer surface
Separate mentions, citations, and recommendations; they are not interchangeable.
Check each item only after the artifact meets the standard. Progress saves on this device; use the academy-home backup to move it elsewhere.