The Complete ClaudeAI Search Department

Tool 01: Question Miner.

Read every call note in project knowledge. Then read what I paste below
from review sites, Reddit or other communities: [paste threads, reviews].

1. Pull every question a buyer asked or implied about choosing a product
   like ours. Quote each one exactly, with its source and date.
2. Mark any question that contains our company name as BRAND. Keep those
   in a separate list. They belong in a brand check, not the benchmark.
3. Mark any question you rewrote or wrote yourself as WRITTEN.
4. Do not pad the list. If there are only 11 real questions, return 11.

Save as question-candidates.md.

Tool 02: Decision Mapper.

Read question-candidates.md and the company facts file.
1. Name the main buying decision our buyers make, as the decision-stage
   term that names our category or product.
2. For each candidate question, say which part of that decision it
   serves: understanding the problem, comparing approaches, building a
   shortlist, choosing between vendors, or checking fit.
3. Flag questions that don't connect to a buying decision at all.

Add a "Decision it serves" column to question-candidates.md.

Tool 03: Intent Sorter.

Read question-candidates.md.
Sort every question into one of four kinds:
- exploring: doesn't know the options yet
- wants names: asking which vendors exist
- wants a verdict: has names, wants a recommendation or comparison
- checking fit: asking whether a product suits their situation

Then mark each question DECIDE or KNOW. A definition question ("what is
X") is usually KNOW. A question with a real buyer constraint ("best X for
a team that has to pass a security review") is usually DECIDE.

Prefer questions with a real constraint. A constrained question tells
you far more than a plain one.
Update question-candidates.md with the kind and DECIDE or KNOW.

Tool 04: Set Builder.

Read question-candidates.md.
1. Propose up to 20 questions for the tracked set. Favor DECIDE
   questions, cover all four kinds, and use the buyer's exact words.
2. For every candidate you did NOT include, say why in one line: kept,
   rewritten, merged, moved or left out.
3. Write a short note to [owner name] asking them to pick the final set,
   and explaining that once locked, the wording won't change.

After the owner replies with their picks, I will paste them here. Then
fill in question-set.md: give each tracked question an ID, set status
to "tracked", record who approved it and today's date.

Tool 05: Run Sheet Builder.

Read question-set.md. Use only questions with status "tracked".
Build this month's run sheet:
1. One row per question per tool: ChatGPT, Claude, Perplexity and Google
   AI Mode. That's four rows per question.
2. For each row: question ID, exact wording, tool, a blank for the date,
   and a blank for the pasted answer.
3. At the top, the run rules:
   - New chat for every question. Never ask two questions in one chat.
   - ChatGPT: a Temporary chat. Claude: an incognito chat. Perplexity: a
     new thread. Google: the AI Mode tab, not the regular results page.
   - Memory and personalization off where the setting exists.
   - Paste the full answer, including the sources shown beside it.
   - Never mention our company in the chat.

Save as run-sheet-[month].md.

Tool 06: Answer Logger.

I'm pasting answers from this month's run sheet below, one at a time or
all together. For each answer, record:
- question ID, tool, date
- named us: yes or no
- every company named, in the order named
- every source shown beside the answer (title and link)
- the exact sentence about us, if we were named

If I paste something that isn't a full answer, ask me for the rest.
Never fill in an answer I haven't pasted.

Save as answer-log-[month].md.

[paste answers]

Tool 07: Answer Grader.

Read answer-log-[month].md and the company facts file.
Grade every answer where we were named:
- POSITION: first, middle or last among the companies named
- ACCURACY: does what it says about us match the company facts file?
  Quote any line that doesn't, next to the correct fact.
- FIT: does it describe us as right for the buyer in the question, or
  for someone else?
- FRAMING: helpful, neutral, or costs us the deal (for example, "only
  for enterprise" when we sell to mid-market)

For answers where we weren't named, note who was named first.
Add every inaccurate description to fix-log.md, with the quote as
evidence. Save the grades in answer-log-[month].md.

Tool 08: Engine Comparer.

Read answer-log-[month].md.
Make a one-screen comparison:
1. For each tool, how many tracked questions named us, out of how many
   were run. Show this month and last month.
2. The company named most often instead of us, per tool.
3. Questions where the tools disagree most: we're named in some and
   missing in others.
4. Any answer that changed a lot from last month, quoted.

Report each tool on its own. Never blend them into one score. The
differences between tools are often the most useful finding.

Tool 09: Robots Checker.

Here is our robots.txt: [paste the full file from yoursite.com/robots.txt].

Check what it allows and blocks for each of these user agents:
- Googlebot (Google Search, including AI Overviews and AI Mode)
- Google-Extended (a separate control for AI training and grounding
  outside Search)
- OAI-SearchBot (ChatGPT search) and GPTBot (OpenAI training)
- Claude-SearchBot (Claude search), Claude-User (fetches pages when a
  user asks) and ClaudeBot (Anthropic training)
- PerplexityBot (Perplexity search)

For each, say: allowed, blocked, or not mentioned (which means allowed
unless a wildcard rule blocks it).
Flag any rule that blocks a search crawler from pages we want found.
Note the difference between blocking training and blocking search, so
we don't block one when we meant the other. The decision on training
bots is a business decision for us to make, not a fix.
Add real problems to fix-log.md.

Tool 10: Render Checker.

For each page below, I'm pasting two things: the text I see in the
browser, and the page source (View Source, not Inspect).
[paste for up to five pages]

For each page:
1. Is the main description of what we do present in the page source as
   text? Or does it only appear after scripts run?
2. Are pricing, integrations, security and FAQ content in the source,
   or loaded later, in tabs, accordions or images?
3. Are headings real headings (H1, H2), and do they say something
   specific?
4. Is any important claim only in an image or a video?

List what a crawler that doesn't run scripts would miss. Add real
problems to fix-log.md.

Tool 11: Index Checker.

For each important page, I'm pasting the Search Console URL Inspection
result: [paste results]. I'm also pasting the Pages report summary:
[paste].

For each page:
1. Is it indexed? If not, what reason does Search Console give?
2. Is the Google-selected canonical the page we expect?
3. Any noindex, nosnippet or max-snippet settings that would stop it
   showing with a snippet?
4. When was it last crawled?

Then list pages from the Pages report that are excluded and look
important. Add real problems to fix-log.md, most serious first.

Tool 12: Schema Checker.

For each page below, I'm pasting the visible text and the JSON-LD
structured data from the page source.
[paste]

For each page:
1. What schema types are present (Organization, Product,
   SoftwareApplication, FAQPage, BreadcrumbList, others)?
2. Does every claim in the schema match the visible text? Quote any
   mismatch.
3. Is anything in the schema that isn't on the page at all?
4. Suggest schema the page could carry, matching the visible text only.

Schema doesn't get you into AI answers on its own. It makes a page
easier for any system to understand, and it has to match what people
see. Add real problems to fix-log.md.

Tool 13: Page Picker.

Read question-set.md, the latest answer log and fix-log.md.
Here is a list of our main pages: [paste URLs and titles].

1. For each tracked question, which of our pages should answer it? If
   none, say "no page".
2. Rank our pages by how many tracked DECIDE questions they should
   answer.
3. Recommend the one or two pages to work on this month, and why.
4. List "no page" questions for Part V.

Improving a few high-value pages usually beats publishing more pages.

Tool 14: Translation Rewriter.

Here is the current text of the page: [paste].
Read the company facts file, the call notes and question-set.md.

1. List every feature the page mentions.
2. For each, write what it lets the buyer do and why that matters to
   them, in words buyers actually used in calls. Quote the call phrase.
3. Mark any feature with no buyer phrase behind it.
4. Rewrite each section with the old copy next to the new copy. Keep
   every fact the same. Only change how clearly it's said.

Mark anything you're unsure about [confirm] for the product owner.

Tool 15: Answer-First Editor.

Here is the page after tool 14: [paste].
1. Quote the current first two sentences.
2. Rewrite them so a new buyer learns what we do, who it's for, and the
   one thing that makes us different, in the buyer's words.
3. Under each heading, make the first sentence answer the heading
   directly. Then add detail.
4. Check the page against the tracked questions it should answer. Mark
   each ANSWERED, PARTLY or MISSING, quoting the line that answers it.

Return the edited page with changes marked.

Tool 16: FAQ Builder.

Read question-set.md, question-candidates.md and the company facts file.
For the page I'm working on: [page name].

1. Pick four to eight real buyer questions this page should answer that
   the body doesn't cover yet.
2. Answer each in one or two plain sentences, using only the company
   facts file. Mark anything missing [confirm].
3. Write FAQPage JSON-LD that matches the visible FAQ word for word.

Never write an FAQ answer that makes a claim we can't support.

Tool 17: Comparison Drafter.

Competitor: [name]. Read the company facts file, the latest answer log
and question-set.md. Here is the competitor's public pricing, product
and security pages: [paste text, with links].

Draft a page:
1. Title in the buyer's words, from a tracked question.
2. The answer in the first two sentences, including who each product
   fits best.
3. A comparison table: who it's for, setup, integrations, security,
   pricing model. Our side only from company facts. Their side only from
   their public pages, each with a link and the date I read it.
4. Four to six FAQs from real buyer questions.
5. At the end: "Facts to confirm" and "Claims to check with legal".

Never write anything about the competitor that their own pages don't
support. Save as compare-[competitor].md.

Tool 18: Alternatives Drafter.

The page: "[Competitor] alternatives". Read the company facts file and
the latest answer log.

1. Write who usually looks for alternatives to [competitor], and why,
   using buyer words from calls and communities where we have them.
2. List the alternatives buyers actually consider, including us,
   based on the answer log and the call notes. For each: who it fits,
   in one or two plain sentences, with a link to a public source.
3. Put us in the list honestly, not automatically first.
4. Add four FAQs from real buyer questions.
5. Add "Facts to confirm" at the end.

Save as alternatives-[competitor].md.

Tool 19: Fair Fit Writer.

Read the draft: [compare or alternatives file].
Write a section called "Where [competitor] is the better fit".
1. Two to four real situations where the competitor is the better
   choice, based on their public pages and our own "who it's not for".
2. Plain and specific. No backhanded compliments.
3. End with one line on who we fit best instead.

A comparison page with no fair section reads like an ad, to buyers and
to anyone summarizing it.

Tool 20: Claim Checker.

You never write new content. You check it.
Read the draft: [file]. Read the company facts file.

1. List every factual claim about us: price, customer, integration,
   certification, feature, result, comparison.
2. Check each against the company facts file. Mark MATCH, NOT IN FACTS
   or WRONG.
3. Check every claim about a competitor has a link and a date.
4. Check nothing promises we'll be named or recommended by any AI tool.
5. Check any FAQ schema matches the visible FAQ word for word.

If everything passes, add "CHECKED [date]" at the top. If anything
fails, add "FIX:" at the top with each problem. The tool that wrote a
draft can't be the one that signs it off.

Tool 21: Source Reader.

Read answer-log-[month].md.
1. List every source that appeared beside an answer. Group by domain.
2. For each domain: how often it appeared, and which companies were
   named in those answers.
3. Mark sources that appear beside competitors but never beside us.
4. Say what each source is: our site, a competitor's site, a review
   profile, an editorial list, a forum thread, a partner directory,
   documentation.

Report these as sources that appeared. Never call a source the reason an
answer named someone.

Tool 22: Mention Finder.

Read the Source Reader output and the company facts file.
For each source that appears three or more times beside a competitor
where we don't appear:
- Review profile: what's missing from ours, filled in from company
  facts, plus a short review ask to a customer who has already offered
- Editorial list: a short note to the author pointing to something real
  and checkable
- Forum or Reddit thread: read the community rules. If vendors may reply
  with disclosure, draft a reply of up to 80 words from a named person,
  who says who they are and where they work in the first line. If the
  rules ban vendors, write "don't post".

Never suggest bought or incentivized reviews, fake accounts, upvotes,
or paid placements that aren't labeled. Add each item to fix-log.md.

Tool 23: Fix Picker.

Read fix-log.md, question-set.md and the latest answer log.
Rank every open fix by:
1. How many tracked DECIDE questions it touches
2. How serious it is (blocked or wrong comes before unclear)
3. How much effort it takes

Return three lists:
- DO NOW: up to five fixes, each with an owner and a reason
- NEXT: real, but not this month, one line each
- NOT DOING: work that looks urgent and isn't, one line each on why

A short list beats a long one. Update the status column in fix-log.md.

Tool 24: Report Writer.

Read the Engine Comparer output, the Fix Picker output and fix-log.md.
Write a one-page report:
1. Lead with the decision we need, not the story.
2. What we checked: how many tracked questions, which four tools, on
   which date.
3. What we found, per tool, this month against last month. No composite
   score.
4. Wrong or costly descriptions of us, quoted.
5. What we shipped last month, and what we're fixing this month.
6. What this can and can't tell us.

Write it so it survives being forwarded to a CEO or CFO who wasn't in
the room. Every number must trace back to the answer log.