# 10 Claude workflows for GEO: the prompts

From A/GEO Agency: https://ageoagency.com/insights/claude-workflows-for-geo/

How to use these prompts:
1. Make a Google Sheet with these tabs: Questions, Facts, Citations, Coverage, Sources, Competitors, Entity.
2. Fill in the context block once. Replace each [PLACEHOLDER] with your own details.
3. For each workflow, paste the context block, then the prompt, into a new scheduled task in Claude.
4. Run each task one time by hand. Check the result. Then set the schedule.

---

## Context block

Fill in this block once. Paste it at the top of each of the 10 prompts, because each scheduled run starts a new session with no memory of the last one.

```text
## Context
- Brand: [BRAND]
- Website: [WEBSITE]
- Category: [CATEGORY]
- Our buyers: [BUYERS]
- Competitors: [COMPETITORS]
- Engines: ChatGPT, Perplexity, Google AI Mode, Claude
- Engine access: [ENGINE ACCESS]
- Log: the Google Sheet "[LOG]". Tabs: Questions, Facts, Citations, Coverage, Sources, Competitors, Entity.
- Questions: the tab "Questions" has one buyer question per row.
- Facts: the tab "Facts" has one true fact about [BRAND] per row: name, address, phone, prices, services, founder and service area.
## Rules
- Do not publish, post, send or upload anything. Write drafts and log rows only.
- Do not change or delete old rows in the log. Add new rows.
- Give the URL for every source that you use.
- Do not guess. If you cannot get an answer, write "no data" and say why.
- Do not invent numbers, quotes, customers or results.
```

---

## 1. AI citation tracker

Asks each engine your buyer questions every day. Logs where AI names or cites you, and where it names a competitor in your place.

- When: Daily
- Needs: Engine access

```text
Run the daily AI citation check.

1. Read the tab "Questions".
2. Ask each engine each question 3 times. AI answers change from run to run, so one run is not enough.
3. For each run, record these fields:
   - date, engine, question, run number (1 to 3)
   - named: "yes" if the answer names [BRAND], else "no"
   - cited: "yes" if the answer links to a page on [WEBSITE], else "no"
   - the [BRAND] URLs that the answer cites
   - all other URLs that the answer cites
   - the competitors that the answer names
4. Add one row per run to the tab "Citations".
5. Calculate today's mention share for [BRAND]: the runs that name [BRAND], divided by all runs. Calculate it for each engine and for all engines. Do the same for each competitor.
6. List the lost questions: questions where the answer names a competitor and does not name [BRAND].
7. End with a summary of 5 lines or less: the total mention share, the best engine, the worst engine, the top 3 lost questions with the competitor that won each one, and each engine that you could not reach.
```

---

## 2. Prompt-coverage check

Takes your buyer questions and flags each question where the AI answer never names you.

- When: Weekly, Monday
- Needs: Workflow 1 log

```text
Run the weekly prompt coverage check.

1. Read the rows from the last 7 days in the tab "Citations".
2. For each question, calculate the mention share for [BRAND] across all engines and runs.
3. Give each question one status:
   - Covered: the answers name [BRAND] in 50% of runs or more.
   - Weak: the answers name [BRAND] in 1% to 49% of runs.
   - Missing: the answers never name [BRAND].
4. For each Weak or Missing question, find the page on [WEBSITE] that answers it best. Use [WEBSITE]/sitemap.xml. If no page answers it, write "no page".
5. Add one row per question to the tab "Coverage": week, question, mention share, status, best page, and the competitor that the answers name most often.
6. End with a summary: the count of Covered, Weak and Missing questions, and the 5 Missing questions that are closest to a decision to buy.
```

---

## 3. Answer-gap finder

Finds the questions about your category that AI tools answer badly. Each gap becomes the brief for your next page or post.

- When: Weekly, Wednesday
- Needs: Web search

```text
Find the questions about [CATEGORY] that AI tools answer badly.

1. Make a list of 20 questions that buyers ask about [CATEGORY] and that are not in the tab "Questions". Use web search results, "People also ask" boxes, Reddit and forum threads, and review sites. Write each question in the words that a buyer uses.
2. Ask 2 engines each question one time.
3. Score each answer from 1 to 5 on each of these 4 tests:
   - Correct: the facts are true.
   - Specific: the answer gives names, numbers and steps, not general advice.
   - Current: the facts are from the last 12 months.
   - Sourced: the answer cites a strong source.
4. Mark an answer as a gap if its total score is 12 or less, or if it has a wrong fact, an old price or no source.
5. For the 5 biggest gaps, write a content brief:
   - Title: the question, in the buyer's words.
   - What the AI answers get wrong or leave out.
   - A draft answer of 40 to 60 words for [BRAND] to check. Mark each claim that needs a source.
   - The proof that we need: data, examples, quotes or our own results.
   - 4 to 6 related questions for an FAQ.
6. Save the 5 briefs in one Google Doc named "Answer gaps [date]".
7. List the other questions that fit what [BRAND] sells. Do not add them to the tab "Questions". I decide which questions to track.
```

---

## 4. Source-authority map

Shows which third-party pages the engines cite for your category, so you know where to get mentioned.

- When: Monthly
- Needs: Workflow 1 log

```text
Map the third-party pages that AI engines cite for [CATEGORY].

1. Read the rows from the last 30 days in the tab "Citations". List each cited URL that is not on [WEBSITE] and not on a competitor's website.
2. Group the URLs by domain. For each domain, count the questions and the engines that cite it.
3. Give each domain one type: review site, directory, list article, forum, video, news, Wikipedia, industry group, or other.
4. For the top 20 domains, open the cited pages. Record whether each page names [BRAND], and which competitors it names.
5. For each top domain that does not name [BRAND], write one next action. Examples: claim a listing, ask the author to review [BRAND] for a list article, answer a forum thread with real help, offer data to a writer, or ask customers for reviews. Do not suggest paid links or fake reviews.
6. Add the results to the tab "Sources" with this month's date: domain, type, citation count, engines, names [BRAND] (yes or no), competitors named, next action.
7. End with the 5 domains where one action can bring the most citations.
```

---

## 5. Entity audit

Checks how AI engines describe your brand, and flags each wrong fact with the place to fix it.

- When: Monthly
- Needs: Engine access

```text
Check how AI engines describe [BRAND].

1. Ask each engine these 6 questions, with no other context:
   - What is [BRAND]?
   - What does [BRAND] sell, and what does it cost?
   - Who is [BRAND] for?
   - Where is [BRAND], and who runs it?
   - What do customers say about [BRAND]?
   - What are the best alternatives to [BRAND]?
2. Compare each answer with the tab "Facts".
3. Mark each claim in each answer as correct, wrong, out of date, or missing. For each wrong claim, copy the exact words and name the engine.
4. For each wrong or out-of-date claim, find the source that the engine cites. If the engine shows no source, write "no source shown".
5. For each problem, write the fix and the place to make it: a page on [WEBSITE], a profile (Google Business Profile, LinkedIn, Crunchbase, review sites, directories), or the schema on the home page.
6. Add the results to the tab "Entity" with today's date.
7. End with the 3 wrong facts that can cost [BRAND] a sale, and the fix for each.
```

---

## 6. llms.txt builder

Writes and checks your llms.txt file, then checks that your brand facts match on every page. Few AI crawlers read llms.txt today, so the fact check is the main value.

- When: Monthly
- Needs: Website access

```text
Build and check the llms.txt file for [WEBSITE].

Note: Few AI crawlers request llms.txt today. Keep this task small. The facts on the pages of [WEBSITE] and in its schema matter more.

1. Read [WEBSITE]/sitemap.xml. Then read the home page, the About page, the pricing page, the service pages and the contact page.
2. Write a draft llms.txt in the format from llmstxt.org:
   - An H1 with the brand name.
   - A summary of one sentence in a blockquote.
   - 5 to 10 short facts: what [BRAND] sells, for whom, where, the prices if they are public, and how to make contact.
   - H2 sections with links. Add only the pages that a buyer needs: services, pricing, proof, FAQ and contact. Give each link one plain sentence.
3. If [WEBSITE]/llms.txt exists, compare it with your draft. List the links that return an error, the pages that are missing, and the facts that changed.
4. Check that these facts match in 4 places: the llms.txt draft, the About page, the Organization schema on the home page, and the tab "Facts". The facts are: name, address, phone, prices, founder and service area. List each mismatch.
5. Save the draft in Google Drive as "llms.txt draft [date]". Do not upload it to the website.
6. End with a summary: what changed since last month, and each mismatch to fix.
```

---

## 7. Competitor-mention pull

Lists the sources that the engines cite for your category, and shows which competitor wins each one.

- When: Weekly, Monday
- Needs: Workflow 1 log

```text
Run the weekly competitor mention report.

1. Read the tab "Citations". Use 2 periods: the last 7 days, and the 7 days before that.
2. For [BRAND] and each competitor, calculate these values for each period:
   - Mention share: the runs that name the brand, divided by all runs.
   - Citation share: the runs that cite the brand's website, divided by all runs.
   - The 5 questions where the answers name the brand most often.
3. For each competitor, list the sources that the engines cite when they name that competitor. Mark each source that does not name [BRAND].
4. Compare the 2 periods. Flag each brand whose mention share went up or down by 10 points or more. Flag each brand that is not in the competitor list and that appears in 3 answers or more.
5. Open the page that won the biggest new citation for a competitor. In 3 sentences, say what the page gives that our page does not: a direct answer, data, a newer date, a comparison, or a strong source.
6. Add the results to the tab "Competitors" with this week's date.
```

---

## 8. Freshness watch

Flags the day that your citation drops out of an answer, with the page that replaced you.

- When: Daily, after workflow 1
- Needs: Workflow 1 log

```text
Run the daily freshness watch. Run this task after the daily AI citation check.

1. Read the tab "Citations" for today and for the 7 days before today.
2. Find each drop. A drop is a question and an engine where both of these are true:
   - The answers named or cited [BRAND] in 50% of runs or more over the 7 days before.
   - The answers did not name or cite [BRAND] in any run today or yesterday.
   Use 2 days, not 1, because AI answers change from day to day.
3. For each drop, record the engine, the question, the [BRAND] URL that the answers cited before, and the URLs that the answers cite now.
4. Open the [BRAND] URL. Check that the page loads, that it does not have a noindex tag, and that robots.txt does not block it. Note the date that the page shows.
5. Open the top URL that replaced it. Note its date and what it answers that our page does not.
6. If there is one drop or more, start your result with "ALERT: [number] citation drops" and list them. If there are no drops, write only "No drops today."
```

---

## 9. Quotable-rewrite pass

Drafts a rewrite of each slipping page, so it gives the clear answer that an engine can quote. You approve every change.

- When: Weekly, Thursday
- Needs: Workflows 1 and 2 logs

```text
Draft rewrites for the pages that lose AI citations.

1. Read the tabs "Coverage" and "Citations". Pick 3 pages on [WEBSITE]. First pick pages with a drop in the last 7 days. Then pick the best pages for Weak questions.
2. For each page, read the full page. Then read the top 2 pages that the engines cite in its place.
3. Write a new draft of the page in this order: answer, explain, prove, expand.
   - Answer: put a direct answer of 40 to 60 words under the H1.
   - Make each H2 a question. Answer it in the first sentence under the heading.
   - Prove: give each main claim a number with its source, its date and its sample. Use only sources that you can link.
   - Where the page needs our own data, write [ADD OUR DATA: what is needed]. Do not invent numbers, quotes, customers or results.
   - Use the names of the brand, the product and the place. Do not use "it" or "this" when the subject is not clear.
   - Expand: add 4 to 6 FAQ questions with answers of 2 to 4 sentences, and FAQPage schema that matches them.
4. Keep the URL, the search intent of the title and the brand voice. Keep every true fact from the old page.
5. Save each draft as a Google Doc named "Rewrite: [page title]". Put a list of the changes at the top.
6. Do not publish. End with the 3 doc links and the question that each rewrite targets.
```

---

## 10. Weekly visibility report

Shows your citation share across the engines and gives you three moves for next week.

- When: Weekly, Monday, last
- Needs: All logs

```text
Write the weekly AI visibility report.

1. Read the tabs "Citations", "Coverage", "Sources", "Competitors" and "Entity".
2. Write a report of one page in a Google Doc named "AI visibility report [date]":
   - First line: one sentence with the change in mention share and the top move. Example: "Mention share went from 18% to 23%. Top move: update the pricing page."
   - A table: mention share and citation share for this week and last week, for each engine and for all engines, for [BRAND] and each competitor.
   - Wins: the questions where the answers name [BRAND] now and did not name it last week.
   - Losses: the questions where [BRAND] dropped out, with the page that replaced it.
   - Open problems: the wrong facts from the tab "Entity" that are not fixed.
   - 3 moves for next week. For each move, name the page or the source, the action, and the question that it targets. Rank the moves by the gain that you expect for the work.
3. Use only numbers from the log. If an engine has no data this week, say so.
```