Skip to content

Understanding Prompt Types

By Matthew Edgar · Last Updated: April 23, 2024

Traditional search is focused on optimizing for queries. Through almost three decades of work, we all have a deeper understanding of search queries and the process involved in optimizing for those.

Conversational search or chat search, powered by generative AI, is focused on optimizing for prompts. We need to make sure that prompts related to our company surface information about our products and services.

How do we optimize for prompts and make sure our companies show up in generative AI search responses? How is this different than optimizing for search queries? It begins by understanding what types of prompts people use.

Prompt Open-Endedness

At the simplest level, one way of grouping prompts is to consider how open-ended the prompt is.

  • Open-ended prompts are vague and contain few details about how a response should be generated. The person entering the prompt is not clear about what type of response they want.
  • Non-open-ended (close-ended) prompts are more detailed and contain specific instructions about how to generate a response. The person entering the prompt knows what type of response they want to see.

With open-ended prompts, the AI tool is not given much detail from the person entering the prompt. Instead, the AI tool must draw upon its training data to infer context and determine what type of response should be provided. The response can go in many different directions and there is more creativity in how the AI tool generates the response. There is typically not a “correct” response, so repeating the same prompt will generate many different types of responses. These types of prompts often lead to follow-up questions and longer conversations with the AI tool.

With non-open-ended prompts, the AI tool can use the prompt to understand the context. The generated output can draw upon more information within the prompt itself or find more specific information in a knowledge base. There is typically a right answer to the question, so repeating the same prompt will typically generate very similar responses. The prompt may also provide more constraints for how the AI tool should generate the response, such as length limits, a particular format or a limit to what information should be used within the generated response.

There are plenty of examples of both types of prompts, with a few presented below (generated with help from Gemini and Claude).

Write a poem about nature.Write a 150 word blog post about baking cookies using chocolate chips, flour, sugar, and butter.
Plan an ideal vacation.Summarize the plot of Harry Potter and the Sorcerer’s Stone in 2 sentences.
Tell a funny story from your childhood.Translate this sentence into Spanish: “Hello, how are you?”
Create a painting that captures the feeling of happiness.Generate a product description for a new smartphone, highlighting its key features and benefits based on the specifications included in the attached PDF. [Attach PDF]
Provide advice for someone feeling stressed.Discuss 5 main causes of global warming in a 250 word essay. Use factual language and avoid sensationalism.

The differences in open-endedness also change intentions. Open-ended prompts are about exploring new ideas and looking for novel responses. The intent of a non-open-ended prompt is to retrieve specific information that is more focused. That makes the intent of a non-open-ended prompt closer to the intentions behind conducting a traditional search, even though the nature of the response will vary greatly from a list of search results.

Prompt types change intentions

Chat Prompts vs. Search Queries

The concept of open-endedness does not apply as easily to search engines. Queries may vary by specificity, ranging from a broad query that could mean many things (“apple”) to a narrow query (“nutritional value of apple”), but search queries are more open-ended in nature than prompts.

Queries are open-ended because no instructions are provided to the search engine about how the search engine should provide results. There is a preset way search results are returned. For example, a search query would never say “summarize the nutritional value of an apple in a table with columns for calories, carbohydrates, and fiber levels.” Even if a query were to add this level of detail, Google’s traditional search engine is not programmed to respond to it while a conversational AI tool would be able to respond. In contrast, a conversational AI tool could respond to this type of prompt (and likely respond well).

There are some examples of non-open-ended search queries but even here they are still more open-ended than a prompt. For example, a search query may focus on a specific brand name or may include a website address. That narrows which websites Google will list in the search results. Still, though, this is pretty open-ended controls over how the response is provided compared to the level of detail that can be specified in a non-open-ended prompt.

Conversation Types

Another way of grouping prompts is by considering the different types of conversations users will have. Research from Nielsen Norman Group found there were six types of conversations users had with conversational AI tools.

  1. Search: This is equivalent to a search query entered on Google or Bing. The prompt entered tends to be short (just like a search query). While the user is looking for specific information, the short prompt does not contain enough context for the AI tool to know how to respond. The AI tool will still attempt to respond, drawing upon its knowledge base to determine the best way to respond. However, the generated response may not satisfy the user. Of course, the same is true for traditional search—the search engine may also misunderstand the user’s intent given the vague nature of the query.

    Where the AI tool can only generate a single response, traditional search results can return a list of many different websites that address this query in a multitude of ways. As a result, for this type of prompt, traditional search results will typically deliver a better experience. This indicates there may be some types of queries that are better suited to traditional search instead of conversational AI and both experiences would be needed. Because of this, both Google and Bing currently incorporate the generated response alongside traditional search results—sometimes the generated response is sufficient, other times the user will need to see a list of results.

    For example, in the following screenshot Copilot’s generated response makes an educated guess the user wants a general summation of the subject matter. The user may have been looking for something else and this generated response might not satisfy the user’s vague prompt. Given that, Bing lets users toggle between Copilot’s response and a traditional search result.
Search prompt types do not contain sufficient information, so the AI tool must make a guess about the user’s intentions
  • Funneling: A funneling conversation starts with a vague and undefined prompt but through refinements in subsequent prompts, the conversation turns toward more specific information. Nielsen Norman Group notes the user has a specific information need but is unsure how to articulate that need. The user can rely on the AI tool to help them refine the prompt through follow-up questions.

    This is dramatically different than traditional search. Google may provide multiple websites in the search results and may offer additional search terms to consider using. However, that may not be enough to help users refine the query. Users will need to spend more time visiting multiple websites and reading multiple articles to determine how best to articulate their information needs.

    In contrast, services like ChatGPT can take the user by the hand and help guide them through the process of better articulating their need. In the following example, ChatGPT responds to a vague prompt offering guidance on ways to make the request more specific. A funneling conversation with a well-trained conversational AI tool would significantly decrease the time required to obtain the desired information compared to hunting through a bunch of websites and search results.
Funneling conversations – the AI tool helps the user better articulate their need
  • Exploring: This type of prompt is equivalent to a conversation with a teacher or a subject matter expert, asking questions to explore an information space. While these conversations often require follow-up questions like a funneling conversation, the follow-up questions are about helping the user deepen their knowledge instead of refining their undefined, poorly articulated prompt.

    With websites and search engines, the user would need to explore their own ideas through their own research, which can increase the time spent obtaining information. Websites might be able to point a user in the right direction. In contrast, a search tool powered by generative AI can ask questions to fully understand exactly what the user needs to help the user explore the most relevant information.

    An exploring conversation can be improved if the conversational AI tool is specifically trained within a specific information space. For example, a chatbot trained on your company’s products could become the most knowledgeable customer service representative your company has ever had: that well-trained chatbot could help customers explore solutions to their problems and find how they can best utilize your company’s products.
Exploring conversations help the user deeply explore a particular subject matter
  • Chiseling: Where exploring conversations are about acquiring a depth of knowledge, chiseling conversations are about the breadth of knowledge. Users are attempting to get a wide array of ideas and perspectives about a given topic. The generative AI tool can also suggest follow-up questions to help the user find ways of considering more aspects of the subject matter being discussed. It is important that we ensure information from our companies is included in these varying perspectives.

    While search engines may be able to surface unique perspectives about a subject, the way search results are presented may make it difficult to know if the website will offer a unique perspective before clicking to it. A lot of the websites ranking for a given query tend to provide much of the same information. So, you would have to click on multiple websites to see if they offer something different. This requires more work compared to a chiseling conversation with a generative AI tool.

    A generative AI tool reduces the work of finding a breadth of information because the user can specifically ask for a variety of perspectives. Also, the AI tool will have been trained on hundreds, thousands or millions of documents related to that subject, allowing it to provide a broader perspective than any user would be able to on their own.
Chiseling conversations are about exploring a breadth of information
  • Pinpointing: A pinpointing prompt is very specific, detailed, and well-articulated. This prompt contains plenty of context for the AI tool to understand how to respond and the user has a specific type of result in mind. Pinpointing conversations involve non-open-ended prompts, limiting the scope of the AI tool’s response.

    This type of prompt requires more work from the user to prepare but can often result in a better response. If the initial prompt is written well enough, then there will be no additional follow-up prompts required. Inevitably, these will be later-stages prompts—the user will likely have had to do additional searching so that they have enough knowledge to craft a detailed and specific prompt.

    There is no equivalent to this type of prompt in traditional search. Traditional search queries are not long enough to provide this level of detail. Plus, traditional search engines are not capable of processing this level of detail. Also, pinpointing prompts tend to be task-oriented, asking the AI tool to generate a specific type of response. Traditional search engines are not designed to perform these types of tasks, so this experience can only be offered by conversational AI.
Pinpointing prompts are very specific, detailed, and well-articulated
  • Expanding: These conversations start with a narrowly focused prompt, but the initial response is too narrow to be satisfactory. Through follow-up prompts, the user expands the conversation further to cover other related topics and questions. The follow-ups are meant to take the conversation in a new direction and not to refine the initial response. During an expanding conversation, the user expects the AI tool to present other details and questions to consider. Expanding conversations will typically start with non-open-ended prompts but will expand with the use of open-ended prompts.

    An expanding conversation could occur with traditional search engines with repeated search queries. However, generative AI search can remember previous prompts and use that memory to improve subsequent responses. That makes it feel more like a conversation, creating a very different experience than search engines or websites can provide. Not surprisingly, every generative AI tool offers a way to easily ask follow-up questions to continue the conversation.
Expanding conversations allow the user to ask follow-up questions that take the conversation in new directions

Key Takeaways

The question for the future of SEO is: are you optimizing your content for all the types of conversations people might have with a chatbot? Chances are that your content is only written for the conversations that mirror search queries. That is how we have all optimized our content over the years.

Optimizing for longer conversations requires providing content that trains the bots on what follow-up questions to ask, provides more depth and breadth about the topic, and trains the generative AI bots on what types of details need to be included in the conversation. Ideally, those details include information about your company.

Optimizing to appear in conversations also requires thinking about new types of experiences our users would prefer. Conversing with a generative AI tool and querying a search engine does not deliver the same type of experience. There is an overlap between these experiences but each is distinct.

Some users will prefer to shift toward conversational AI tools for some types of conversations. For example, non-open-ended conversations allow the user more control over the response than they can get from traditional search engines. Generative AI tools also make it easier to ask follow-up questions to seek a breadth of information, a depth of information, or take the conversation in new directions—none of that is easily doable within traditional search engines.

Users seeking specific information may still prefer to use traditional search engines. Users entering open-ended prompts may find traditional search results a preferable response compared to how generative AI would respond. In some cases, a conversation isn’t needed—the user simply wants to find a website that addresses their specific need.

We need to assess the queries currently driving traffic to our website and assess what type of response our users would prefer. Would our users still prefer to search for all those queries and find our website in search results? Or would users prefer to converse with a generative AI tool instead? For any queries where users prefer to converse, we need to understand how they will converse and see if information about our company shows up in those conversations.

Want Help Approaching SGE?

Do you want help assessing your search queries and finding the types of experiences your users prefer? We can help you do the necessary research and formulate a generative AI search strategy. Contact me for help.

You may also like

What Are Generative AI Hallucinations?

One of the biggest problems with generative AI responses is hallucinations. What are hallucinations? Why do they happen? More importantly, as generative AI starts to dominate search results, how do we factor hallucinations into our future SEO strategies?

Will People Click Links in SGE Results?

Will people click links in SGE results? How many clicks could we lose with SGE rolling out? We conducted an in-depth study to find out.

Search Intents & Generative AI

How will SGE change SEO? It likely depends on the search query intent. Learn how to adapt your SEO strategy and prepare for SGE.