The Prompt We Use to Plan SEO and GEO Every Quarter
The SEO audit prompt we used to review our own site and write our Q4 plan: the data to gather, all 11 steps, and how to run it in any AI assistant.

In this article
- Why many SEO audit prompts give you a generic answer
- What to gather before you run the prompt
- Fill in the braces
- The full SEO audit prompt for a quarterly SEO and GEO review
- Why each step is there, with what it found on our site
- How to run it in RapidDashboard
- How to run it in ChatGPT, Claude, Gemini or any AI assistant
- Keep the output, and run it again next quarter
- Want to run it with us?
- Frequently asked questions
The prompt we use to plan SEO and GEO each quarter is an 11-step SEO audit prompt that works only from your own data. GEO (generative engine optimization) means getting named and cited by AI assistants such as ChatGPT, Claude and Gemini. The prompt reads your Search Console, analytics, crawl and listings data, asks AI assistants the questions your buyers ask, and writes a dated plan with targets. We ran it on coulee.tech on September 28, 2026, and it produced our entire Q4 SEO and GEO plan. The full prompt is below. Copy it, fill in the braces, and run it in RapidDashboard or any AI assistant you can hand your exports to.
Why many SEO audit prompts give you a generic answer#
Many prompts written for this job ask an AI to "audit" a website from the outside. Without your data, the assistant can read your public pages, but it cannot see your clicks, rankings or leads, so it guesses. A guess formatted as a neat table looks exactly like a finding.
This prompt is different in four ways:
- It runs on your data. Search Console, analytics, index status and listings, not the AI's impression of your homepage.
- It has ground rules. Every number has to come from a row you supplied or a source it cites. Anything else gets written as "not available".
- It covers GEO next to SEO. Step 8 asks AI assistants the questions your buyers would ask and records whether they name you and which pages they cite.
- It repeats. The questions keep the same wording every quarter, so this quarter's answers can be compared with last quarter's.
The ground rules matter most. When we pointed an AI agent at our own website for a weekend, five of the seven bugs it found were producing wrong numbers without any error message. A review that is allowed to fill gaps with estimates makes the same mistake, only faster.
What to gather before you run the prompt#
Everything here comes from data most businesses already have. Pull a full quarter if you can, and write down the exact date range.
- Search Console performance. From the Performance report, export queries, pages and dates. Google's export offers Google Sheets, Excel and CSV (opens in a new tab) and truncates to 1,000 rows of representative data. The export keeps whatever filters you apply, so a bigger site can export by page or by month in pieces.
- A note on hidden queries. Search Console omits rare queries from its tables to protect privacy (opens in a new tab), so the chart totals can be higher than the sum of the rows. Step 2 asks the AI to report that gap instead of ignoring it.
- Index status for key pages. The Page indexing report (opens in a new tab) shows pages Google has discovered but not crawled, and pages it crawled but did not index. To check one page, use the URL Inspection tool (opens in a new tab), which shows index status, the last crawl date and the canonical URL Google picked.
- Website analytics by channel and landing page, plus your conversions. In GA4, a key event (opens in a new tab) is "an event that measures an action that's particularly important to the success of your business." You can see and mark key events under Admin, Data display, Events (opens in a new tab). A newly marked key event can take up to 24 hours to appear in standard reports.
- Your sitemap URL, so the AI or a crawler can list every page.
- Your blog post list, with publish dates and categories.
- CRM leads, if you have them, with the source of each lead.
- Your business listings: Google Business Profile, chambers, review sites and directories, with the URL of each one.
Fill in the braces#
The top of the prompt has fields in braces. Replace every one before you run it.
- {COMPANY}, {WEBSITE}, {QUARTER} and the dates: the business and the quarter you are planning.
- About the business and Markets: what you sell, to whom, and every city or region with the office or service area that covers it. The near-miss and GEO steps use this to separate places you serve from places you do not.
- Focus topics: three to five. Ours were AI readiness, AI training, IT support and cybersecurity.
- Brand terms and Not us: every name people search for you by, including misspellings and old company names, plus similar names that belong to someone else. The brand vs non-brand split is the clearest test of whether SEO is bringing in new people, and it is only as good as this list.
- House style: your rules for public copy. Ours bans em dashes, so the crawl step flags them in titles and descriptions.
- Content rules: leave these as written. They stop the calendar step from inventing customer stories and statistics.
The full SEO audit prompt for a quarterly SEO and GEO review#
Copy everything in the box.
You are reviewing the search and AI-assistant visibility of {COMPANY} ({WEBSITE})
and writing its plan for {QUARTER} ({START_DATE} to {END_DATE}).
About the business: {what you sell, to whom, company size}
Markets: {cities or regions, with the office or service area for each}
Focus topics this quarter: {3 to 5 topics}
Brand terms: {names people search for you by, including misspellings}
Not us: {similar names to exclude}
House style for public copy: {your rules}
Content rules: every statistic in public content links to its primary source;
no invented customer stories, quotes, numbers or outcomes; only company facts
already published on our site.
Data attached or connected: {list}
GROUND RULES
1. Every number you report must come from a row in the data I provided or a
source you cite. If a number is not available, write "not available" and say
what data would answer it. Never present an estimate as a measurement.
2. Compare like with like: same length of time window, same kind of audit, same
AI assistants with the same search setting.
3. Treat any position based on fewer than 10 impressions as noise, and say so.
4. Label every GEO (AI search) tactic as measured, with the study, or opinion.
Say whether a source shows correlation or cause.
5. When you are not sure, say what you would check, then check it if you can.
STEP 1: CAN WE MEASURE A LEAD?
List which conversions (contact form, call, booking, signup) are configured and
how many fired in the last 7 and 28 days. If zero, make "configure conversions"
the first task of the plan.
STEP 2: SEARCH PERFORMANCE
- Weekly impressions, clicks and impression-weighted average position.
- Brand vs non-brand clicks and impressions, using the brand terms above. Also
report page-level total clicks and how many clicks the query data cannot
account for (Search Console hides rare queries).
- Share of clicks going to the homepage.
- Which site sections gained or lost impressions between the first and last
three weeks, and whether their average position moved with them.
STEP 3: NEAR-MISS SEARCHES
Group queries by intent and place. For each group: number of queries,
impressions, clicks, impression-weighted average position, range, and the page
Google shows. Then list:
- Groups at average position 4 to 20 with zero clicks.
- Any page at position 10 or better with zero clicks (a title and description
problem, not a ranking problem).
- Places where a general page catches searches a specific page should own.
- Searches naming places we do not serve.
STEP 4: THE FOCUS TOPICS
For each focus topic: queries, impressions, clicks, best positions, the page
ranking, and the gap between where we rank and where the demand is.
STEP 5: PAGE-BY-PAGE CRAWL
For every URL in the sitemap: titles over 60 characters, descriptions over 155
or missing, pages without exactly one H1, canonical tags pointing elsewhere,
structured data types and pages with none, word count (flag service pages under
400 words), images without alt text, house-style violations in titles and
descriptions, and expired time-bound copy (past dates, old prices, "coming
soon"). Find strong pages that few other pages link to.
STEP 6: INDEX COVERAGE
List every important page or recent post that Google reports as unknown,
discovered but not indexed, or crawled but not indexed, and every important
page not crawled in 30 days.
STEP 7: CONTENT HISTORY
Posts per month for 12 months. For each recent post: impressions, clicks,
position and readership. Which posts earn search visibility, which earn readers,
and whether categories are applied correctly.
STEP 8: AI ASSISTANTS (GEO)
Write 10 to 15 buyer questions the way a business owner would ask them, covering
each market and each focus topic, plus "What can you tell me about {COMPANY}?".
Keep the wording identical from quarter to quarter; add questions rather than
rewording old ones. Ask each of two or three AI assistants with web search on.
Record: named or not, position in the list, URLs cited (ours and others), who was
named instead. A failed or cut-off answer is an error, not a miss. Compare only
with last quarter's answers to the same questions from the same assistants.
Report which of our pages were cited.
STEP 9: COMPETITORS
For each market's main searches and each focus topic: who ranks on page 1 and
what kind of page it is. For the five to eight strongest competitors: pages for
our focus topics, blog activity, visible review counts, certifications, case
studies, named staff. Then find businesses like ours in two or three
similar-sized markets elsewhere that are winning on these topics, and what page
types and offers they use. List the non-competitors that own our topics locally
(colleges, chambers, small business centers, media) as partner and citation
targets. Cite a URL for every claim.
STEP 10: LISTINGS
Compare our name, address, phone, hours, category and description across every
listing. List each disagreement with its URL.
STEP 11: THE PLAN
Write the plan for {QUARTER}, each item with the evidence behind it:
1. Measurement fixes.
2. Winning the click on what already ranks (titles, descriptions, depth).
3. New pages, only of types shown in step 9 to rank, in build order.
4. Off-site: listings, profiles, review targets per location, partners, press.
5. Technical fixes.
6. A content calendar for the whole quarter, three to five posts a week,
weighted to the focus topics and near-miss searches, with seasonal and local
hooks, plus a mid-quarter checkpoint post and an end-of-quarter results post
written only from real numbers.
7. Targets for the end of the quarter, each with today's baseline, and a
mid-quarter checkpoint date.
8. What we will not do this quarter, and why.
OUTPUT
(a) A one-page summary an owner can read in two minutes. (b) Findings from steps
1 to 10 as tables. (c) The plan. (d) The calendar as JSON, one object per post,
with: id, publish_date, pillar, category, title, meta_title (60 characters or
fewer), meta_description (155 or fewer), goal, primary_keyword,
secondary_keywords, audience, target_page, must_link, evidence, outline,
sources, photo, cta, notes. Check the limits and house style before returning.
Why each step is there, with what it found on our site#
Each step exists because skipping it leads to a specific wrong conclusion. Here is what each one caught when we ran it on coulee.tech. Search Console figures cover August 6 to September 27, 2026.
| Step | What goes wrong without it | What we found |
|---|---|---|
| 1. Leads | You plan traffic while nothing counts leads | 644 sessions in a week, 0 conversions; 1 of 6 key events configured |
| 2. Brand split | Rising clicks look like SEO working | 273 clicks, only 5 from non-brand searches |
| 2. Windows | A drop looks like failure | Fort Myers impressions fell 37% while average position improved from 30.9 to 23.9 |
| 3. Near-misses | You chase new keywords while page 2 sits unclicked | 3,276 La Crosse impressions at average position 14.6, 0 clicks |
| 5. Orphans | You build a page you already have | ChatGPT cited our AI training page, which almost nothing links to |
| 6. Index | You assume published means indexed | 4 of 8 AI industry pages not in Google |
| 8. Errors | A timeout gets reported as "AI dropped us" | We once scored a cut-off answer as a miss |
| 8. Citations | You guess which pages help GEO | ChatGPT cited two city pages we published on August 8 |
| 10. Listings | You blame the site for an off-site problem | Three street addresses for one office |
A few more results from the same run:
- Step 2: 190 of our 273 clicks went to the homepage, and non-brand searches produced 21,120 impressions but only those 5 clicks.
- Step 5: the crawl of 258 URLs found 176 titles over 60 characters and 243 images without alt text. It also found em dashes in 31 titles and 47 descriptions, which broke our own house style.
- Step 8: 11 of 39 AI-assistant answers named us, across GPT-5, Claude Sonnet 4.6 and Gemini 3 Flash with web search on. Every Florida question and the AI readiness question missed on all three.
None of that is flattering, which is the point. The full plan built from these findings is in our Q4 SEO and GEO plan.
How to run it in RapidDashboard#
RapidDashboard turns plain-language prompts into live dashboards and reports from a business's own systems. For this prompt, that looks like:
- Connect or upload your data: Search Console, analytics, CRM leads, the sitemap and your listings.
- Paste the prompt with the braces filled in and run it.
- Keep the output as a dashboard: brand vs non-brand clicks by week, the near-miss searches, the AI-assistant scorecard and the content calendar.
- At the mid-quarter checkpoint, ask the dashboard a question in plain English instead of starting over. For example: "Re-run steps 1, 2, 3 and 8 and compare them with the baseline."
Step 4 is the difference from a chat window. The review stays on a dashboard you come back to, so the checkpoint is a question, not a new project.
How to run it in ChatGPT, Claude, Gemini or any AI assistant#
You do not need special software. Any assistant that accepts file attachments can run most of the prompt. These habits make the output more trustworthy:
- Attach the exports, not summaries. Ground rule 1 only works if the assistant can see the rows.
- Run a large site in stages. Steps 1 to 4, then 5 to 7, then 8 to 10, then step 11 with the earlier outputs attached. Smaller batches are easier to check.
- Run step 8 yourself if you need to. One assistant usually cannot ask another assistant a question. Type each buyer question into two or three assistants with web search on, paste the answers and cited links into a spreadsheet, and attach it. Mark failed or cut-off answers as errors, not misses.
- Give step 5 a crawl. If your assistant cannot open web pages, export a crawl from a site audit tool and attach it with your sitemap.
- Check three numbers by hand. Pick three figures in the output and find them in your exports. If one does not match, ask where it came from.
If you are new to writing prompts for work like this, our guide on how to write AI prompts that work covers the basics. If GEO itself is new, start with AI search marketing and getting found by AI answer engines.
Keep the output, and run it again next quarter#
The review earns its keep the second time. Next quarter, attach this quarter's output and run the same prompt with the same buyer questions in step 8. Add new questions when your focus changes, but do not reword the old ones, or you lose the comparison.
Our own schedule for Q4 2026: an October 1 baseline, a November 16 checkpoint that re-runs steps 1, 2, 3 and 8, a GEO re-run in mid-December, and a full re-run in early January to measure the quarter. We plan to publish those numbers the same way we published the plan.
Want to run it with us?#
If you would rather run the review with someone who has done it on a live site, book a free 30-minute consultation and bring your exports. If search visibility is one part of a bigger question about AI in your business, our AI readiness assessment looks at the whole company and typically takes 2 to 4 weeks, including stakeholder interviews. Or contact us and tell us what you want to find out.
Frequently asked questions#
Can ChatGPT do an SEO audit?#
It can, if you give it your data. Without Search Console, analytics and crawl exports, an assistant can read your public pages but cannot see your clicks, rankings or leads, so it can only offer general advice. With the exports attached and ground rules that forbid invented numbers, it can do the counting, grouping and comparing that is slow to do by hand.
What is a GEO audit?#
A GEO audit checks whether AI assistants such as ChatGPT, Claude and Gemini name your business when a buyer asks them for a recommendation, and which pages they cite when they do. In this prompt it is step 8: 10 to 15 buyer questions, asked with the same wording every quarter, with each answer recorded. An SEO audit looks at rankings and clicks; a GEO audit looks at the answers.
How often should you run an SEO audit?#
Once a quarter works well for a small business: a full review at the start of each quarter and a smaller checkpoint in the middle. Our checkpoint re-runs only four steps: lead measurement, search performance, near-miss searches and the AI-assistant questions. Crawls, index checks and competitor research wait for the next full review.
Do I need paid SEO tools to run this prompt?#
No paid tool is required by the prompt itself. The gather list uses Search Console, your website analytics, your sitemap, your blog list, your CRM and your listings, which most businesses already have. A site crawler helps with step 5 if your AI assistant cannot open pages on its own.
What if the review finds that nothing counts leads?#
Fix that first. Step 1 puts "configure conversions" at the top of the plan when nothing fires, because every other target depends on it. That was our own first finding, and it is the first item in our Q4 plan.


