“Best X for Y” Prompts: How AI Assistants Build Recommendation Lists

When someone asks an AI assistant for “the best X for Y”, it usually searches the web, reads a handful of pages such as comparison articles, review sites, directories and forum threads, and then names the brands that come up most often and most credibly for that specific need. Your odds of being listed depend less on your own homepage and more on whether trusted pages about your category include you, and say clearly who you are best for.

How a recommendation answer is put together

No AI company publishes its exact method, and each engine works differently. Based on what the answers and their sources show, the process usually looks like this:

1. Interpret the question. The engine pulls out the category (“project management software”), the constraint (“for a construction company”) and any context like budget or location.
2. Search. It runs one or more web searches, often several variations of the question.
3. Read sources. It reads a set of pages: “best of” articles, review platforms, directories, forums such as Reddit, and sometimes brand websites.
4. Find agreement. Brands that appear across several trusted sources, and that are described as fitting the specific need, are the strongest candidates.
5. Write the answer. It lists a few options, often with a short reason for each and links to its sources.

Answers based only on training data skip the search step and rely on what the model already learned, which favours brands that were widely written about before the model’s cutoff.

What makes a brand more likely to be listed

1. Presence on the pages engines read

If the three most-cited “best CRM for agencies” articles do not mention you, it is hard for an engine to recommend you for that question. See which websites ChatGPT cites and how to get onto them.

2. A clear fit for the “Y”

Engines match brands to needs. A page that says “built for construction teams of 10 to 50 people” gives the engine a reason to pick you for that prompt. A page that says “the all-in-one solution for everyone” does not.

3. Consistent facts

Price, location, features and audience should match across your site and profiles. Conflicting details make engines hedge or skip you.

4. Reviews and reputation

Ratings and review summaries often show up in recommendation answers, especially for local and consumer categories.

5. Readable pages

If AI search crawlers are blocked or your key information is hidden in images or scripts, engines cannot use it. Check your robots.txt.

Common “best X for Y” prompt patterns

Worked example: why one brand made the list and another did not

Two bookkeeping firms in the same city offer almost the same service. For the prompt “best bookkeeper for restaurants in Manchester”, one is named in most answers and the other in almost none. Looking at the sources explains it:

The second firm may be just as good. The engine simply has no evidence that it fits the “for restaurants” part of the question.

Illustrative example.

Mistakes that keep brands off AI lists

• Generic positioning. “Solutions for everyone” matches no specific prompt.
• Hidden pricing. For “under $X” prompts, engines cannot include what they cannot price.
• Unclaimed profiles. Empty or outdated directory and review profiles in your category.
• No comparison content. If competitors publish honest comparison pages and you do not, their framing becomes the source.
• Blocked crawlers. A single robots.txt rule can remove you from an engine’s search results.

How to find out where you stand

1. List 10 to 20 “best X for Y” prompts your customers really ask. Our guide to prompt research vs keyword research shows how to build the list.
2. Ask each in ChatGPT, Perplexity, Gemini and Claude with web search on, two or three times.
3. Record who is listed, in what order, and every cited source.
4. For each source, check whether it mentions you, a competitor, or neither.
5. Work on the gaps: sources that list competitors but not you, and use cases where your positioning is unclear.

Onxeera’s AI prompt research tool runs these prompts across engines, records brands, positions and sources, and highlights the prompts where competitors are named and you are not. If a competitor keeps winning, read why ChatGPT recommends your competitor instead of you.

Frequently asked questions

Can I pay to be included in ChatGPT recommendations?

Not in the organic answers. Some AI platforms have started testing ads, but those are labelled separately from the answer. Being recommended in the answer itself depends on what the engine finds and trusts, not on payment.

Why does the list change every time I ask?

AI answers are generated fresh each time, and engines may retrieve different pages on different runs. Brands with strong, consistent signals tend to appear in most runs; brands on the edge appear only sometimes. Measure across several answers.

Does the order of the list matter?

Often, yes. Readers pay most attention to the first few names, and some answers explicitly label a top pick. Position is noisy from run to run, so track your typical position rather than one result.

Do reviews influence AI recommendations?

Reviews and ratings often appear in the sources engines cite for recommendation questions, and engines sometimes summarise them in the answer. A steady flow of genuine reviews on the platforms engines read in your market helps.