SEO teams spend hours on tasks that are 80% pattern recognition: clustering keywords, writing meta descriptions, spotting broken links, comparing competitor headings. That's exactly the kind of work LLMs handle well — if you give them structured prompts instead of vague requests. Below are 12 prompts I actually use, each with a usage example, expected output, and the task it replaces.
A quick rule before we start: LLMs hallucinate. Search volumes, backlink counts, and ranking positions must come from real tools (Google Search Console, Ahrefs, Semrush, Serpstat). Prompts are for structuring, analyzing, and generating — not for inventing data. Feed them real exports.
Semantic Core and Keyword Clustering
1. Building a semantic core from a business description
Task: replace the first pass of keyword research when you have no seed list yet.
You are an SEO strategist. Business: [description].
Target region: [country/city]. Language: [language].
Generate a semantic core of 80-120 queries grouped by:
- informational (how, what, why)
- commercial (buy, price, order)
- navigational (brand, login)
- transactional (download, sign up, book)
For each query give: intent, estimated funnel stage (TOFU/MOFU/BOFU), and a suggested landing page type.
Output as a Markdown table. Do not invent search volumes.
Example request: "Business: online course platform for IT professionals. Region: global, English. Generate the core."
Expected result: a table of 80+ queries with intent labels, grouped by funnel stage. You then validate volumes in a keyword tool and drop the noise.
2. Clustering a raw keyword export
Task: group thousands of queries into topical clusters without manual sorting.
Here is a CSV of queries with volumes: [paste].
Cluster them by SERP intent (queries that would rank on the same page).
Rules:
- max 15 clusters
- each cluster gets a name, a primary keyword, and a list of secondary keywords
- flag clusters with fewer than 3 queries as 'thin'
Output: cluster name
| primary keyword | total volume | secondary keywords.
Why it works: LLMs are good at semantic similarity. Always spot-check by searching 2-3 queries per cluster in Google — if the SERPs differ, split the cluster.
3. Identifying content gaps against competitors
Task: replace a manual competitor content audit.
I have heading structures (H1-H3) from 3 competitor pages: [paste].
My page covers: [paste my headings].
Compare and output:
1. Topics competitors cover that I miss (with the competitor name)
2. Topics I cover that none of them do (my differentiators)
3. Suggested new H2 sections in priority order with a one-line rationale each.
Expected result: a prioritized list of missing subtopics. This is the fastest way to find content gaps before writing.
Content: Meta Tags, Briefs, and Strategy
4. Generating title and meta description variants
Task: produce 10 compliant meta tag pairs per page in one pass.
Page topic: [topic]. Primary keyword: [keyword].
Generate 10 variants of <title> (max 60 chars) and <meta description> (max 155 chars).
Rules:
- keyword appears naturally in the first 40 characters of the title
- no clickbait, no ALL CAPS
- each variant uses a different angle (benefit, number, question, year, comparison)
Output as a table: #
| title | length | description | length.
Expected result: a ready-to-pick table. Character counts matter — Google truncates titles around 580 pixels, roughly 60 characters.
5. Writing a content brief for a writer
Task: replace the brief-writing hour before every article.
Create a content brief for the query: [primary keyword].
Include:
- target audience and search intent
- 5-7 H2 sections with 2-3 bullet points each
- questions from 'People Also Ask' to answer (I'll paste them: [paste])
- internal link targets (I'll list my relevant pages: [paste])
- word count range based on top-3 competitors: [paste their word counts]
- 3 unique angles competitors miss
Why it works: the brief becomes a checklist, and the writer (human or AI) can't drift off-topic.
6. Building a 3-month content plan
Task: turn a keyword list into a publishing calendar.
Here are my clusters with volumes: [paste].
Build a 12-week content plan.
Rules:
- prioritize clusters by (volume x commercial intent) / competition
- 1 pillar page + 3 supporting articles per cluster
- schedule pillar pages first, supporting articles 1-2 weeks apart
- output: week
| article title | target keyword | type (pillar/supporting) | internal links to add
Expected result: a calendar you can paste into Notion or a spreadsheet. Review priorities manually — LLMs overvalue high-volume informational queries.
Technical Audit
7. Parsing a Screaming Frog export for issues
Task: replace the first triage of a crawl export.
Here is a Screaming Frog CSV export with columns: URL, Status Code, Title, Meta Description, H1, Word Count, Response Time. [paste]
Find and report:
1. Pages with status 4xx/5xx and their count
2. Duplicate titles and duplicate meta descriptions (list URLs)
3. Missing or multiple H1s
4. Pages with word count < 300 (thin content)
5. Pages with response time > 1s
Prioritize by potential traffic impact. Output as a numbered action list.
Expected result: a prioritized fix list in seconds. This is the single highest-leverage prompt for technical SEO audits.
8. Writing redirect map for a site migration
Task: build a 301 redirect map from old to new URLs.
Old URLs: [paste list]. New URLs: [paste list].
Match each old URL to the most relevant new URL based on slug similarity and topic.
Output a table: old URL
| new URL | match confidence (high/medium/low) | reason.
Flag any old URL without a good match for manual review.
Expected result: a draft redirect map. Manually review all "low confidence" matches — a wrong 301 kills rankings.
9. Auditing structured data
Task: validate and improve Schema.org markup.
Here is my JSON-LD: [paste].
Check it against Schema.org requirements for [Article/Product/FAQPage].
Report:
1. Missing required properties
2. Missing recommended properties
3. Syntax errors
4. Suggested additions to improve rich result eligibility
Reference exact property names from schema.org.
Expected result: a checklist of fixes. Always verify against the official Schema.org documentation and Google's Rich Results Test.
Analysis and Reporting
10. Decoding Search Console query data
Task: find opportunities in a GSC export.
Here is a Google Search Console export: Query, Clicks, Impressions, CTR, Position. [paste]
Identify:
1. Queries with high impressions but CTR < 2% (title/meta opportunity)
2. Queries ranking 4-10 (quick wins — small push to top 3)
3. Queries with position 11-20 (content expansion candidates)
4. Branded vs non-branded split
Output 3 prioritized actions with expected impact.
Expected result: a short action list. The "position 4-10" bucket is usually the fastest traffic gain.
11. Analyzing SERP intent for a target query
Task: decide what content format Google rewards.
For the query "[query]", here are the top 10 results: [paste titles, URLs, content types].
Analyze:
- dominant content type (guide, listicle, product, video, tool)
- average word count
- common subtopics
- whether featured snippet / PAA / video block appears
Recommend the format and structure my page should use to compete.
Expected result: a format recommendation backed by SERP evidence. Better than guessing.
12. Turning a monthly report into plain language
Task: replace the reporting hour for stakeholders.
Here is my monthly SEO data: [paste metrics].
Write an executive summary for a non-technical stakeholder.
Structure:
- 3 wins (with numbers)
- 2 concerns (with cause and fix)
- next month's 3 priorities
Keep it under 200 words. No jargon.
Expected result: a report your client or manager actually reads.
What Actually Changes
Used together, these prompts cut the routine part of SEO work: clustering, meta generation, crawl triage, brief writing. What remains is the part that matters — validating data, choosing priorities, and making decisions the model can't.
Start with one prompt, test it on real data, refine the wording until the output is usable. The best prompt library is the one you actually iterate on. If you want to go deeper into AI-assisted workflows, the ASI Biont blog has more hands-on material on prompts and automation.
Comments