Keywords vs. Prompts: What Marketing Teams Need to Measure Now

For about 20 years, SEO success meant one thing: getting your page into the top ten results for a keyword. AI tools like ChatGPT, Perplexity, Gemini and Google AI Overviews work differently. People now type long, detailed questions, and the AI writes an answer instead of showing a list of links.

This means marketing teams now need to track two things: keyword rankings and how often AI tools recommend their brand.

Keywords and prompts are not the same
A keyword is a short search phrase of two to four words, like "ecommerce customer service software." It shows what someone wants but gives little context. Google matches the words to pages and returns a ranked list of links.

A prompt is a full question with details. For example: "We have three Shopify stores and 15 support agents, and we use Klaviyo. We need helpdesk software with a flat monthly price. Which tools should we look at?"
A keyword asks a search engine to find pages. A prompt asks an AI to compare options and recommend specific brands.

Why prompts need a different kind of tracking
Prompt results change because AI models like GPT-4o, Claude and Gemini build each answer word by word based on probability. If you run the same prompt five times, you may get different brands.
AI engines also do not search for the whole prompt at once. Google's documentation on AI Overviews describes a process called query fan-out. The AI splits the prompt into several smaller searches, reads the pages it finds, pulls out the useful facts, and combines them into one answer.
So there is no single rank to track. Instead, you run each prompt many times and measure how often your brand shows up.

ut. Getting listeExample: one buyer, two kinds of search
Take OmniDesk, a made-up support platform for online stores. It costs a flat $199 per month for 20 agents, routes tickets across several Shopify stores, and connects with Klaviyo.

An ecommerce manager wants to replace their ticketing tool.
On Google, they type "shopify multi store customer service tool." Google shows ten links. OmniDesk ranks third and gets clicks.
In ChatGPT or Perplexity, they write a longer request: three Shopify stores, 15 agents, uses Klaviyo, needs multi-store ticket routing and a flat price with no per-ticket fees. Behind the scenes, the AI runs three smaller searches:
• Shopify multi store helpdesk ticket routing
• ecommerce helpdesk Klaviyo integration
• customer support software flat pricing no per-ticket fees
It reads review sites and buying guides, drops any tool that charges per ticket, and recommends three platforms.
Now suppose OmniDesk only shares its price after a demo request. The AI cannot confirm the price, so it recommends competitors whose prices are listed openly. OmniDesk wins on Google but is left out of the AI answer.

Four metrics for tracking prompts

1. Share of answers: the percentage of runs where the AI recommends your brand. If a prompt is run 30 times and your brand appears 18 times, your share is 60%.
2. Prominence: whether you are the main pick or a backup option, and whether the AI describes your product correctly.
3. Competitor mentions: which competitors show up for the same prompts, and which requirements make the AI choose them.
4. Cited sources: the pages the AI uses as evidence, such as review roundups, product docs and forums. Cited sources are the most useful of the four. When a competitor gets recommended, the citations show which articles leave you od on those sites can change future answers, because AI engines update their sources regularly.

Keywords still matter
Prompts do not replace SEO. You need both.
First, AI tools depend on search engines. OpenAI's documentation on OAI-SearchBot shows that AI search assistants pull source pages from search indexes like Bing and Google. If your pages have crawl errors or do not rank for the smaller searches, the AI cannot find or cite them.
Second, people still use short keywords for some searches. Someone looking for "OmniDesk login" types a keyword, not a prompt.
Keyword tracking tells you whether people can find your website. Prompt tracking tells you whether AI tools recommend your brand. Our guide on traditional SEO vs GEO and AEO explains how the two work together.

How to start tracking prompts

1. Collect real customer questions. Pull 20 questions from sales calls and support tickets. Include details like team size, budget and the tools they already use.
2. Test them every week. Run your prompts in ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews to find your starting share of answers. Note which competitors appear.
3. Check the sites AI tools cite. Find the sources cited most often in your category. Make sure your brand details are correct there, and reach out where you are missing.
4. Put key facts in plain HTML. AI crawlers often do not run JavaScript, so show pricing and integrations in plain HTML. Allow OAI-SearchBot, PerplexityBot and Googlebot in your robots.txt file. See our guides on how generative engines choose brands and 5 steps to improve your AI rank for the technical details.
5. Keep tracking keywords. Watch your rankings for brand searches and high-intent searches.
6. Automate the testing. Running dozens of prompts by hand takes too long. Tools like Lumirank test prompts daily across six AI engines and track share of answers, competitor mentions and citations.