Build ONE buyer persona from the information below. Give me one named person, not a
segment. Treat this person and their employer as an illustrative hypothesis, not a
verified customer. Label the invented name and employer as illustrative.

MY COMPANY
Name: [OFFICIAL COMPANY NAME]
Website: [YOUR URL]
What we sell: [ONE SENTENCE]
Category and region: [FOR EXAMPLE: china cnc machining service, selling to Europe and the US]
Who we sell to: [YOUR BUYER, BE SPECIFIC ABOUT THEIR JOB AND THEIR COUNTRY]

MY HOMEPAGE TEXT
"""
[PASTE YOUR HOMEPAGE TEXT]
"""

MY BUYER PERSONA
- Their name, job title and the company they work at.
- What they are under pressure to deliver.
- What happens to them personally if they choose the wrong supplier. Write a complete
  consequence sentence: "If they choose the wrong supplier, ..." Give a concrete
  consequence for this person, not a list of buying criteria.
- WHAT THIS PERSON MUST BELIEVE BEFORE THEY CAN SAY YES: 5 to 7 complete statements.
  These are beliefs to test, not claims that my company already meets them.
Do not give me demographic criteria, a list of adjectives or more than one persona.

Keep the supporting company evidence separate, as FACT SHEET, for the later audit:
Fact | Evidence label | Where it came from | Where a buyer could verify it
- One company fact per row, as a complete short sentence.
- Apply the three evidence labels. My homepage and form are CLAIMED, not VERIFIED.
- If the page contradicts itself, mark the row CONFLICTED and quote both claims.
  Do not pick one or average them.
- List FACTS I AM MISSING. Use [VERIFY: what to check] for missing company facts.
- Never put the illustrative persona or inferred buyer beliefs into the Fact Sheet.
Remember MY BUYER PERSONA and FACT SHEET for the rest of this conversation. Every later
claim about my company must trace to the Fact Sheet, with its evidence label intact.

Use MY BUYER PERSONA and my corrected FACT SHEET from this chat. Give me two parts.

PART A. DEAL-BREAKERS
Give me exactly 5. A deal-breaker is an absolute must-have. Fail one and the deal is
dead. It is not a preference and not an objection.
Table: Deal-breaker | Why it kills the deal | Content hole risk (high, medium, low)
Reason from this person's pressure and personal consequence. Include must-haves I did
not mention, but label inferred requirements as hypotheses to check with a real buyer.
Content hole risk means the risk that no public page answers this requirement about my
company. Judge each row separately from the evidence available. The risks must vary.
Give a short reason with each risk rating. Do not invent a public search or claim a
missing page is proven missing when only the homepage was supplied. If evidence cannot
support varied ratings, say they are provisional and name what would change each one.

PART B. MY TEN QUESTIONS
Table: # | Question | Family | What it tests
Exactly 10 questions, each assigned to one of three families:
- At least 4 SHORTLIST questions. This is the moment a buyer picks a supplier:
  "best X in Y" or "who should i use for Z". Make them specific to this buyer.
- At least 3 DEALBREAKER questions, drawn straight from Part A.
- Exactly 1 BRANDED question using my real company name. Never [Company Name].
- The remaining questions are free choice within SHORTLIST or DEALBREAKER.
Exactly 9 questions must be unbranded. My company name and its aliases appear nowhere
in those questions. Never use "you", "your" or "we" in any question. A stranger is typing.
Write lowercase, natural questions, the way a person types into a chatbot. Never
keyword-stuff. No two questions may test the same thing.
Show the count at the end: total, SHORTLIST, DEALBREAKER, BRANDED and unbranded. Check
that the three family counts add to 10, with 9 unbranded and 1 branded.
End with one honest sentence about the biggest blind spot in this map.
Remember DEAL-BREAKERS and MY TEN QUESTIONS for the rest of this conversation.

Open a PRIVATE or INCOGNITO window. Stay LOGGED OUT.
Ask each of your ten questions in a FRESH chat. Never use a follow-up for the next question.
Paste the question and the source-request line below together in the first message.

For each answer, record four things on worksheet page 2:
1. Were you named at all? Yes or no.
2. If the answer cites sources, WHICH OF YOUR OWN PAGES was cited? Write the exact URL,
   not just the domain. Write "none" if no owned page was cited.
3. Which outside websites were cited? Write the domains, or "none".
4. Which competitors were named? Write their names, or "none".

Item 2 is the one almost nobody records. It tells you which page is doing the work.
Record only citations shown in the answer. A guessed link is not a citation.
If you are short of time, eight of the ten questions is enough. Mark the two not run.
Save the answers with their question numbers. Copy your single worst answer in full:
the most wrong, not the most negative. You need its exact text for the audit.

Paste this line together with each question:

List the sources used for this answer with their exact page URLs; if no sources were used, say so.

Use MY BUYER PERSONA, MY FACT SHEET, MY DEAL-BREAKERS, MY TEN QUESTIONS and the answers I
recorded in this chat. If I paste a block headed SCAN EVIDENCE, treat it as measured data
and use it ahead of anything either of us remembers.

Turn the citation test into a content and distribution plan I can start this week.

Three rules for the whole answer:
1. Do not assume that copying a cited page makes an AI cite me. Every recommendation is a
   hypothesis to test, so say what result would prove it wrong.
2. Separate what was OBSERVED in my records from what you INFER. Label the inference.
3. Never invent a citation, a statistic, a certificate or a customer. If the evidence for
   a claim is not in my FACT SHEET, write [VERIFY: evidence needed] in its place.

PART A. WHAT IS ACTUALLY GETTING CITED

One row per outside page that was cited or retrieved in the test.

Question # | Page or domain | Cited or only retrieved | Website type | Page type |
The buyer question it answers | What that page has that mine does not

Website type, choose one: manufacturer or supplier, independent publication, marketplace,
trade association, standards or certification body, engineering resource, comparison or
review site, distributor, university or government, forum or community, other.

Page type, choose one: buyer guide, technical explainer, comparison or list, FAQ, quality
or compliance page, case study, specification or reference page, calculator or tool,
inspection or documentation example, service page, other.

Describe only what is observable on the page. Do not claim to know how the model ranked it.

PART B. THE PATTERN

Across those pages, give:
- the 3 to 5 page formats that appear most often
- the 3 to 5 website types that appear most often
- what those pages do that a generic service page does not
- whether the answers favour narrow buyer-question pages, broad guides, comparisons,
  evidence pages, or something else

If I recorded fewer than five cited pages, say the sample is too small to call a pattern,
and give the strongest single observation instead.

PART C. WHAT TO PUBLISH ON MY OWN SITE

Exactly 8 pieces. Titles specific enough that a writer could start today.

Priority | Page or article | The question or deal-breaker it answers | Format |
Evidence I must have before publishing | Why it could close the citation gap

Start with the questions where I was absent or where an outside page was cited instead.
Do not give me eight blog posts. Include evidence assets where they fit: a sample
inspection report, a material certificate, a process capability example, a comparison
table, an engineering checklist, a case study, a method page, a calculator.

PART D. WHERE ELSE MY INFORMATION SHOULD EXIST

Priority | Domain type | Why a buyer or an AI answer would use it | What to publish there |
How to earn the placement honestly

Allowed: specialist engineering and manufacturing publications, industry directories,
relevant marketplaces, standards and certification registers, supplier databases,
association and conference sites, joint case studies with a named customer or vendor.

Never recommend: fake independent review sites, a site I control that pretends to be
independent, undisclosed paid endorsement, invented reviews, guest-post networks, the same
article spread across many domains, or any fabricated test, statistic or certificate. If
you suggest a comparison on somebody else's publication, require editorial independence
and disclosure.

PART E. BUILD, EARN, LIST

Sort every recommendation into one bucket:
BUILD, content and evidence on my own domain.
EARN, independent coverage, citations, reviews, case studies or certifications that a
third party has to choose to give me.
LIST, accurate factual presence in legitimate databases, marketplaces, directories and
certification records.

Then give the five highest-priority actions across all three, in order, with the reason
each one is above the next.

PART F. THREE BRIEFS

For the top 3 BUILD items:
- the buyer question it answers
- the title
- the page structure, section by section
- the facts I must confirm first
- the tables, diagrams, examples or downloadable evidence it needs
- what would make it more useful than the page that is cited today
- the claims I must not make until they are verified

END WITH THREE LINES
1. My single biggest content gap.
2. My single biggest third-party authority gap.
3. The one page I should publish first, and the one measurement that will tell me in six
   weeks whether it worked.

Keep every [VERIFY: evidence needed] label exactly where you put it.

MY TEN QUESTIONS AND THE ANSWERS I RECORDED
"""
[PASTE YOUR TEN QUESTIONS AND WHAT YOU RECORDED FOR EACH ANSWER]
"""

SCAN EVIDENCE
"""
[PASTE THE SCAN EVIDENCE BLOCK YOU COPIED IN STEP 5]
"""

MY CORRECTED FACT SHEET
"""
[PASTE YOUR CORRECTED FACT SHEET]
"""

Use MY FACT SHEET from this chat. If I paste a block headed SCAN EVIDENCE, treat it as
measured data.

Audit what the public web makes a machine believe about MY BRAND, the company that sells
WHAT I SELL from WHERE I AM. I want the mistakes. Do not write me a company summary.

Five rules for the whole answer:
1. Report only things that could make an AI, a search engine or a buyer get me wrong.
2. My own website is not proof. A claim repeated on five sites is one claim until you find
   where it started.
3. Keep WHAT THE WEB SAYS, WHAT IS PROBABLY TRUE and WHAT AN AI WOULD INFER apart. They
   are three different things and you must label which one you are writing.
4. Never invent a source, a company, a date or a number. Write [NOT FOUND] instead.
5. Short. Tables and single lines. No essays, no restating the question.

PART A. SAME NAME, DIFFERENT THING (up to 6 rows)

Name | What it is | Where | Why a machine could merge it with me | Strength of its record

Strength is high when the other entity owns a Wikipedia page, a Wikidata item, a company
register entry, a patent, an app store listing or major press, because those are the
records a search engine trusts first.

PART B. WRONG OR OUT OF DATE FACTS (up to 8 rows)

Claim on the web | Where it appears | What is probably true | Wrong, outdated, or
technically true but misleading | Chance an AI repeats it, high medium or low

Check at least: founder, owner, parent company, headquarters, country of origin, year
founded, size, category, products, markets served, pricing, and whether anything says I
am closed, acquired, discontinued or unsupported.

PART C. WHAT MAKES AN AI NOT RECOMMEND ME (up to 6 rows)

A statement can be completely true and still cost me the recommendation.

Statement | Source | Who benefits from it | True, outdated or unfair | What a buyer
concludes

Look for: local only, too small, unproven, beginner only, legacy, expensive, hard to work
with, missing a feature I now have, slow support, quality or trust complaints, and any
comparison or alternatives page written by a competitor.

PART D. WHERE THE DAMAGING CLAIM STARTED (top 3 only)

For the three most damaging claims, trace first source, then who copied it, then who
repeats it now. If the trail cannot be established, say so in one line and stop.

PART E. WHAT THE MACHINE PROBABLY BELIEVES

Ten lines, this exact shape, each with one label:

my brand -> attribute -> value [correct | weakly supported | ambiguous | wrong | outdated
| confused with another entity]

PART F. DO NOT ASSOCIATE, DO NOT DESCRIBE AS

Two short lists, only where you found real evidence that the confusion is plausible.
For every do not describe as line, give the description that should replace it.

PART G. THE FIX LIST

Exactly 5 actions, most valuable first, one line each:

Action | Where it goes, my site, my structured data, or a third party | Which problem
above it removes

One of the five must be the single page on my own site that settles who I am, and you
must say what that page has to state.

END WITH TWO LINES
1. The most damaging thing the web currently makes a machine believe about me.
2. The one sentence I should use everywhere to describe myself: name, category, location,
   market served.

MY FACT SHEET
"""
[PASTE YOUR CORRECTED FACT SHEET, OR LEAVE THIS OUT IF IT IS ALREADY IN THIS CHAT]
"""

SCAN EVIDENCE
"""
[PASTE THE SCAN EVIDENCE BLOCK FROM STEP 5, IF YOU HAVE ONE]
"""

You are an on-page auditor for AI search. Your job is to check one web page against
the checklist below and report what to change, so that an AI answer can name my
company for one buyer question. You are an auditor, not a marketer.

======================= MY FOUR INPUTS =======================
BUYER QUESTION: [the question my buyer types, for example: cheapest CRM for a small business]
MY OFFICIAL NAME: [the one name I use everywhere, for example: Acme CRM]
PAGE ADDRESS: [https://...]
PAGE TEXT: [paste the full visible text, or fetch the page if you can.
Say whether you fetched it. Never claim to read a page you did not fetch.]
OPTIONAL, MY FACT SHEET OR TRUTH NOTEBOOK: [paste it if you have one]
OPTIONAL, COMPETITORS THE AI NAMES FOR THIS QUESTION: [names, if I know them]
==============================================================

RULES. Keep the audit honest.

1. LABEL EVERY FACT. Every company fact needs an evidence label. VERIFIED means an
   outside source confirms it. CLAIMED means only my company says it. UNKNOWN means
   evidence is missing. Never upgrade a label without evidence. Rewrites follow the
   separate label layout in report section 4.
2. NEVER INVENT. Do not guess prices, dates, certificates, customers or specifications.
   Use [VERIFY: what to check] wherever a needed fact is missing.
3. USE PLAIN ENGLISH. Use short sentences and everyday words. Define unfamiliar terms
   in six words or fewer. No em dashes, en dashes or marketing adjectives.
4. DESCRIBE GENERAL BEHAVIOUR. Explain what AI search engines generally do.
   Never claim to know what one named engine does today.
5. USE ONLY THESE CHECKS. Do not add checks, link-building advice, word targets or
   an overall numerical score. Record a PASS and move on when a check passes.
6. STOP WITHOUT PAGE CONTENT. If PAGE TEXT is empty and you cannot fetch it,
   ask for the text and stop. If a homepage or blog has no facts about what I sell,
   ask for a product, service or pricing page instead.
7. QUOTE THE PAGE. For every FAIL, quote the exact failing words or write
   "not on the page". If you cannot inspect something, use NOT OBSERVABLE.
   Missing access to evidence is not proof that a check fails.

STEP 0. Map the buyer's needs to your facts.

An ENTITY is a thing with a name. An ATTRIBUTE is a property the buyer asks about.
A VALUE is your answer for that property. A TRIPLE puts all three in one sentence.
An entity class means the kind of thing wanted. CRM means customer relationship
management software. A Truth Notebook is your checked company fact sheet.

a. Split BUYER QUESTION into its parts. Name the entity class, each requested
   attribute and the desired value. For "cheapest CRM for a small business",
   the class is CRM software. Price should be lowest; customer size should be small.
b. Suggest five likely searches and the attribute each explores. Label them as
   suggestions, not observed searches. Consider unstated needs: price unit, free
   plan, minimum order, delivery time, proof and current information.
   Include only needs relevant to this buyer question.
c. Find my value for every attribute, including those in the suggested searches.
   Quote the sentence exactly. Use FOUND when one sentence contains the full triple.
   Use WEAK when the value is vague, split up or missing a unit.
   Use MISSING when the page gives no answer.

Run BLOCKS A to G. Give each check PASS, FAIL or NOT OBSERVABLE, a short reason
and a page quote. PASS requires evidence for the whole condition.
For NOT OBSERVABLE, name the missing access or evidence and how to check it.

BLOCK A. Make clear who the page describes.
A1. Test: the official name identifies the page.
    PASS if its exact spelling appears in the title, main heading and first sentence.
    FAIL if any is missing or different. Fix: use MY OFFICIAL NAME in all three.
A2. Test: the page has one main entity.
    PASS if it describes one product, service or company; FAIL if unrelated offers mix.
    Fix: give each main entity its own page.
A3. Test: the page states what the entity is.
    PASS if the opening names its class beside its name; FAIL if readers must guess.
    Fix: write "[Name] is [product or service type] for [buyer]."
A4. Test: the name agrees across pages and public profiles.
    PASS if spelling matches and old or other names appear once; FAIL if names vary.
    Fix: choose one official name and explain other names once.
A5. Test: key fact sentences name their subject.
    PASS if each starts with the entity name; FAIL if it starts with "we" or "it".
    Fix: replace unclear openings with "[Official name] offers ..." or similar.
A6. Test: readers can identify the publisher.
    PASS if the page is on my domain and names an author or company with an about link.
    FAIL if either is absent. Fix: use my domain and add the publisher and link.

BLOCK B. Answer each buyer need with a value.
B1. Test: the step 0 map has answers.
    PASS if every attribute has a value or clear "does not apply"; FAIL if any is missing.
    Fix: add one factual sentence for each missing answer.
B2. Test: values are specific.
    PASS for numbers with units, dates, named plans or standards, and yes/no answers.
    FAIL for vague words such as "fast". Fix: give the value and its conditions.
B3. Test: each triple fits in one sentence.
    PASS if entity, attribute, value and conditions appear together; FAIL if separated.
    Fix: write "[Name] costs [price] per [unit] on [plan], [billing condition]."
B4. Test: "cheapest", "fastest" and similar claims have support.
    PASS with named comparisons, figures and a check date, or no such claim.
    FAIL without support. Fix: add a dated comparison or remove the claim.
B5. Test: each value includes its conditions.
    PASS if its sentence includes relevant plan, currency, region, material or limits.
    FAIL if needed context is elsewhere or absent. Fix: put conditions beside the value.
B6. Test: the main value is easy to identify.
    PASS with one clear main figure and consistent mentions; FAIL if figures conflict.
    Fix: state the main figure once and explain any different plans or billing periods.
B7. Test: likely follow-up questions have answers here or one click away.
    PASS if relevant costs, trials, contracts, setup, support or factory details appear.
    FAIL if needed answers are absent. Fix: add or link the missing triples.

BLOCK C. Write facts that make sense alone.
C1. Test: each key sentence works outside the page.
    PASS if a stranger understands the subject, fact and unit; FAIL if context is needed.
    Fix: add the name, unit and essential conditions to the sentence.
C2. Test: key sentences are short and focused.
    PASS if each gives one main fact with its conditions; FAIL if separate facts pile up.
    Fix: split separate facts into separate sentences.
C3. Test: related facts form a short, clearly headed passage.
    PASS with two to four sentences under an entity-and-attribute heading; FAIL otherwise.
    Fix: group the facts under a heading such as "Acme CRM costs less for small teams".
C4. Test: key facts appear in readable body text.
    PASS if visible without opening images, files, tabs or tools; FAIL if hidden there.
    Fix: repeat the fact as a plain sentence in the page body.
C5. Test: multiple values are easy to compare.
    PASS with a list or table whose entries name the product or plan; FAIL otherwise.
    Fix: give each entry its own name, attribute, value and conditions.
C6. Test: the key sentence stays consistent across my materials.
    PASS if wording matches my fact sheet and public profiles; FAIL if wording differs.
    Fix: choose one accurate sentence and reuse it. Update every copy when facts change.

BLOCK D. Put the buyer's answer near the top.
D1. Test: the title identifies the entity and buyer's attribute.
    PASS if plain, specific and unique on my site; FAIL if vague or duplicated.
    Fix: name the entity and answer, for example "Acme CRM starts at [price]".
D2. Test: the opening gives the answer promptly.
    PASS if the first two or three sentences name the entity, class and main value.
    FAIL if slogans or welcomes come first. Fix: move the triple to the opening.
D3. Test: the page address names its subject.
    PASS for readable words such as /acme-crm-pricing; FAIL for unclear codes.
    Fix: use a clear address and redirect the old address to it.
D4. Test: labels and values are close and easy to find.
    PASS if key terms appear in the title and once in bold, beside their values.
    FAIL if separated or hard to find. Fix: bold the triple once.
D5. Test: a link can open each attribute section directly.
    PASS if each has a heading and anchor, a direct link target; FAIL if either is absent.
    Fix: add a heading ID, for example <h2 id="pricing">.

BLOCK E. Support facts and keep them current.
E1. Test: the page shows a real update date.
    PASS if the date matches a confirmed edit; FAIL if absent or known to be false.
    Fix: add the actual update date. Change it only when the facts change.
E2. Test: key claims have nearby proof.
    PASS with relevant certificates, test results, captioned photos or outside sources.
    FAIL without proof. Fix: add it, get permission for customer names, or label CLAIMED.
E3. Test: wording matches the evidence.
    PASS if claims reflect their actual status; FAIL if an application becomes approval.
    Fix: use the accurate verb, such as "applied for" instead of "holds".
E4. Test: an outside page repeats the same key fact.
    PASS if a checked page I do not own uses the same words; FAIL if confirmed absent.
    Fix: name a profile to update, or use [VERIFY: which profile]. Check who supplied it.
E5. Test: values agree across my pages and public profiles.
    PASS if checked values agree; FAIL if prices, specifications or conditions conflict.
    Fix: list the conflicting pages and correct each using the confirmed fact.

BLOCK F. Make the page readable by machines.
F1. Test: an AI assistant that can browse is able to fetch this page and quote it.
    PASS if you fetched the page and its quotes and numbers match the text; FAIL if a fetch failed.
    Fix: investigate access. If you could not test it, use NOT OBSERVABLE and ask for the
    access test on pack page 4.
F2. Test: key facts exist in the HTML, the page's source code, before any script runs.
    PASS if they are visible in the fetched source; FAIL if they appear only after loading.
    Fix: put the facts in the source. With pasted text only, use NOT OBSERVABLE and ask
    for the source test on pack page 4.
F3. Test: code uses real headings, lists and tables, with labels beside values.
    PASS for suitable h1, h2, h3, ul, ol and table elements; FAIL for missing structure.
    Fix: ask a developer to replace visually styled blocks with suitable elements.
F4. Test: schema markup, machine-readable fact labels, matches visible facts.
    PASS if suitable types and name, price, currency and unit agree; FAIL otherwise.
    Fix: add or correct Organization, Product, Service, Offer or FAQPage as appropriate.
    Check at validator.schema.org. Markup helps interpretation; it cannot guarantee links.
F5. Test: images and videos have text versions of their facts.
    PASS with suitable image descriptions, captions, transcripts and body text.
    FAIL if facts appear only in media. Fix: repeat the entity, attribute and value in text.

BLOCK G. Remove wording that weakens your facts.
G1. Test: wording is useful rather than repetitive.
    PASS without forced keywords or copied filler across pages; FAIL when either appears.
    Fix: use terms where needed and remove repeated blocks.
G2. Test: uncertainty reflects real limits.
    PASS if qualifiers such as "typically" are justified; FAIL if they hide missing facts.
    Fix: state the confirmed value or a supported range with its conditions.
G3. Test: every section serves the page's sales topic.
    PASS if sections help this buyer assess the offer; FAIL for unrelated traffic topics.
    Fix: remove unrelated sections or move them to a suitable page.
G4. Test: ranges have clear endpoints and units.
    PASS for "from 0.01 mm to 0.05 mm, depending on material"; FAIL for unclear ranges.
    Fix: label both endpoints and state the buyer's applicable value separately.
G5. Test: content avoids unsupported or offensive claims.
    PASS without unqualified medical advice, unsupported competitor claims or abuse.
    FAIL if present. Fix: remove abuse; remove or properly support and qualify claims.

REPORT FORMAT. Show what to change and why.

# Make this page easier to quote
Page address: [PAGE ADDRESS]
Buyer question: [BUYER QUESTION]
Page text: fetched by me / pasted by you (say which)

## 1. Match your facts to buyer needs
| Words in the question | Entity, attribute or value | My value on the page now (quote) | FOUND / WEAK / MISSING |
| ... | ... | ... | ... |
Entity class: [...]
Suggested searches, not observed searches, and the attribute each explores:
1. ...

## 2. See which blocks need work
| Block | PASS | FAIL | NOT OBSERVABLE |
| A. Entity | | | |
| B. Attribute and value | | | |
| C. Standalone facts | | | |
| D. Position | | | |
| E. Proof and date | | | |
| F. Readable | | | |
| G. Conflicting wording | | | |

## 3. Fix the biggest barriers first
List the five most useful fixes, or fewer if fewer checks fail.
Put access failures first, then missing buyer answers, unclear facts and missing proof.
Do not invent a score or recommend changes to passing checks.
1. [Action]. Check [ID]. Why it matters: [...].
   On the page now: "[quote]" or "not on the page".
   Evidence label: VERIFIED / CLAIMED / UNKNOWN.
2. ...

## 4. Paste-ready rewrites for the top three fixes
### [New heading that names the entity and the attribute]
[Two to four short sentences, written as text ready to publish on the page. The first
one is the triple. Use only facts from the page or from my inputs. Where a fact is
missing, write [VERIFY: ...] inside the sentence. Do not write the evidence labels
inside these sentences; a reader of the page must never see the words VERIFIED,
CLAIMED or UNKNOWN. Put the labels on the line below instead.]
Labels: [one label per fact used in the passage, in order]
Intended quote: "[the one sentence I want the AI to lift]"
Ready to publish: [YES, or NO until the [VERIFY: ...] markers above are resolved]

## 5. Confirm these facts before publishing
- [VERIFY: ...]

## 6. See why each check passed or failed
[A1 to G5: PASS / FAIL / NOT OBSERVABLE, one sentence, the quote]

## 7. Check results after making changes
- These checks describe general AI search behaviour, not a named engine's current system.
- A page must be fetched before it can be cited. Test access if F1 is NOT OBSERVABLE.
- Outside pages also supply answers. E4 starts that work; repeated company claims
  do not become independent proof.
- Clear facts and matching markup do not guarantee rankings or links.