How to Convert SEO Keywords Into AI Prompts in Google Sheets

Your keyword list is not obsolete because AI search exists. It is the starting inventory.

That is the useful idea behind a Google Sheets workflow Lily Ray shared on LinkedIn. Lily, listed by Amsive as VP, SEO and AI Search and by Algorythmic as its founder, uses the =AI() or =Gemini() function in Google Sheets to turn rows of terse SEO keywords into natural-language prompts a person might ask ChatGPT, Gemini, AI Mode, or a voice assistant.

The move is simple. Export keywords from Ahrefs, Semrush, Similarweb, Google Search Console, or paid search. Put them in a sheet. Use Gemini in the next column to rewrite each keyword as a conversational AI prompt while preserving the original intent.

The part I like is the order of operations. It does not start with a blank LLM prompt and ask the model to invent demand. It starts with keyword rows that already have some evidence behind them, then translates those rows into prompt-shaped language.

Why SEO Keywords Need A Prompt Translation Layer

SEO keyword data is compact by design. A keyword export gives you strings like:

• women's running shoes
• dj headphones for techno music
• travel to berlin
• berlin travel tips

Those rows make sense in SEO tools. They are easy to sort, filter, cluster, and map to URLs. They do not match how a person talks to an AI assistant.

The same demand sounds more like:

• What are the best women's running shoes?
• What are the best DJ headphones for techno music?
• What is the best way to travel to Berlin?
• What are some travel tips for Berlin?

That translation matters for two reasons.

First, AI visibility tracking usually happens at prompt level. If you are testing ChatGPT, Gemini, Perplexity, Claude, or Google AI Mode, you need full questions or requests, not keyword shorthand. A prompt tracker built from raw keywords will feel artificial.

Second, Google says AI Overviews and AI Mode may use query fan-out, issuing multiple related searches across subtopics and data sources to develop a response. A short keyword can become several retrieval paths once an AI system tries to answer it. That does not make keyword data useless. It means the keyword is the seed, not the whole research object.

I covered the retrieval side in query augmentation and agentic search. This spreadsheet workflow gives you a practical way to prepare the front end of that process: turn keyword inventory into questions worth testing.

The Google Sheets Function

Google's help documentation says Sheets users with access can enter an AI function directly in a cell. The function can be written as =AI() or =Gemini(), and it can reference another cell such as A2.

A simple version looks like this:

=AI("Rewrite this SEO keyword as one natural-language AI assistant question. Preserve intent, brand, product, and location. Do not answer it. Keyword:", A2)

Lily Ray's version is more controlled:

=AI("You are converting a terse SEO search keyword into the natural-language query a real person would actually type or speak to an AI assistant like ChatGPT, Gemini, or a voice assistant. Rewrite the keyword as a single conversational question or request that preserves the exact search intent and intent type (informational, commercial, transactional, or local). Phrase it the way someone would genuinely ask out loud in a full sentence - not in keyword shorthand. Keep any brand, product, or location named in the keyword, but do NOT invent specifics, constraints, or details that aren't already implied. Do not answer the query. Return only the rewritten prompt as plain text - no quotation marks, no preamble, no explanation, no trailing punctuation beyond a question mark. Keyword:", A2)

The prompt has four useful controls:

1. Preserve intent. The output should keep the search job intact.
2. Preserve intent type. Informational, commercial, transactional, and local prompts need different pages.
3. Preserve entities. Brand, product, and location names should not disappear.
4. Do not invent details. The model should not add price, audience, date, or use-case constraints that the keyword did not imply.

That last rule is the one I would watch most closely. If the source keyword is women's running shoes, the converted prompt should not become What are the best women's running shoes for flat feet under $100? unless the original row already implied flat feet and price.

The Sheet Structure I Would Use

Do not stop at two columns. The function is useful, but the workflow gets stronger when the generated prompt stays attached to SEO and business data.

This setup keeps the output grounded. You are not building a random list of AI questions. You are building a prompt inventory connected to keyword volume, page ownership, and business priority.

The Prompt QA Rules

Raw AI-generated prompt rows are not clean data. They need review before you use them for AI visibility prompt tracking, content planning, or reporting.

Here is the QA pass I would run:

• Invented specifics: Flag any prompt that adds a price, audience, date, comparison, or modifier not present in the keyword.
• Intent drift: Check whether commercial keywords became informational, local keywords became generic, or transactional keywords became buying guides.
• Entity loss: Make sure brand names, product names, and locations survive the rewrite.
• Answer leakage: Delete rows where Gemini answers the query instead of rewriting it.
• Duplicate prompts: Watch for different keywords collapsing into the same question. Sometimes that means one page can serve both. Sometimes it means the model flattened useful nuance.

How To Use The Prompt Set

Once the rows pass QA, the prompt set can do four jobs.

1. Build A Better AI Visibility Tracker

Take the clean prompts and test them in the systems that matter for your category: Google AI Overviews, Google AI Mode, ChatGPT, Gemini, Perplexity, Claude, or Copilot.

For each prompt, track:

• whether your brand appears;
• whether your site is cited;
• which URL gets cited;
• which competitors appear;
• whether the answer is accurate;
• whether the answer points users toward a next step.

This is cleaner than starting with invented prompts because each row has a keyword source behind it.

2. Map Content Gaps

Compare prompts against your existing URLs. Some prompts will map cleanly to pages you already have. Others will reveal missing sections, weak comparison coverage, or pages that answer the keyword but not the full question.

For example, a prompt like What is the best way to migrate a Shopify store without losing SEO traffic? may need a service page, a case study, and a process article working together. A prompt like What is hreflang? may need a concise explainer or tool page.

3. Refresh Existing Pages

Prompt rows are useful for page refreshes. If many prompts ask about pricing, comparisons, implementation order, local fit, or enterprise use cases, you can add sections to pages that already have authority.

This fits Google's AI search guidance. Google says there are no special AI search requirements beyond normal SEO fundamentals, but it also says AI features may issue related searches across subtopics. Clear sections, visible text, accurate structured data, internal links, and helpful media still matter.

That is the same point I made in Google's AI search optimization documentation: do the normal SEO work well, then make sure the page answers the connected questions a model may need.

4. Build Internal Links Around User Jobs

Keyword clusters show topic overlap. Prompt clusters show user jobs.

If several prompts move from definition to comparison to vendor selection, your internal links should follow that path. Link from educational pages to comparison pages. Link from comparison pages to service or product pages. Link tactical posts to tools when a tool helps the reader act.

For this topic, the natural path might be:

• prompt tracker methodology;
• Google AI impression reporting;
• an AI search split scorecard;
• a service page for AI SEO strategy.

The Google Sheets Limits To Plan Around

The workflow scales, but it is not infinite.

Google's Workspace Updates post says that when users select multiple cells with AI functions, only the first 200 selected cells with AI functions are generated. You can wait for that generation to finish, then select more cells.

For a small keyword set, this is fine. For a 5,000-row export, build a batching habit:

1. Filter to one topic cluster.
2. Generate 200 rows at a time.
3. Run QA before moving to the next cluster.
4. Keep a status column so you know which rows are generated, reviewed, and accepted.

That sounds slower than one giant bulk generation pass, but it produces better data. You catch prompt drift early instead of cleaning 5,000 messy rows later.

What This Workflow Does Not Prove

This is the part I would be careful about in client reporting.

The generated prompts are not user logs. They are model-generated approximations grounded in keyword demand. They are useful hypotheses, not proof of what people typed into ChatGPT.

The workflow also does not create a direct path into AI Overviews or AI Mode. Google's official guidance says there are no extra technical requirements, no special AI markup, and no unique schema needed for those features. Pages still need to be indexable, eligible for snippets, helpful, technically accessible, and supported by normal SEO fundamentals.

So use the sheet for research and planning. Do not sell it as an AI ranking button.

A Practical Workflow

If I were running this for a client, I would use this order:

1. Export keywords. Pull keyword rows with volume, intent, current rank, current URL, and topic cluster.
2. Convert to prompts. Use Lily's formula or a tighter variant in Google Sheets.
3. Batch generation. Run 200 rows at a time.
4. QA output. Mark each prompt PASS, FLAG, or DELETE.
5. Group rows. Cluster by intent, funnel stage, URL, and business priority.
6. Assign actions. Track, refresh, create, merge, or ignore.
7. Measure separately. Keep keyword rank, AI citation, brand mention, AI impression, and conversion reporting in separate columns.

The final step matters. A keyword rank is not an AI citation. An AI mention is not a click. An AI impression is not revenue. The sheet should help you separate those signals, not mash them into one fake certainty score.

The Practical Takeaway

Lily Ray's formula is useful because it puts AI-search research back on a data footing.

Start with real keyword demand. Convert those rows into conversational prompts. QA the output. Group by intent and funnel stage. Then use the clean prompt set for AI visibility tracking, content gap analysis, page refreshes, and internal link planning.

AI search does not make keyword research irrelevant. It makes the translation layer more important.

About the Author

Francisco Leon de Vivero is VP of Growth at Growing Search, with 15+ years of SEO experience across ecommerce, adult, international, and enterprise search. He worked as SEO Lead at Shopify for more than seven years and spent more than four years working on SEO for Pornhub.

LinkedIn · YouTube · Get in touch