Find literary agents with ChatGPT or Claude
Copy the prompt below into ChatGPT, Claude or Gemini. It sends them to Phrasemill's public pages to pick out the tags that describe your book, then to find the agents who say they want those things.
Let the model read the book
The best results come from attaching your manuscript to the chat and letting the model read the whole thing before it goes looking. It picks up tone, structure and preoccupations that a one-paragraph synopsis flattens, and those are exactly what agents describe when they say what they want. If you would rather not upload the book, the second prompt takes a description instead.
Your manuscript stays in your chatbot. Phrasemill never sees it, and the prompt tells the model not to put any of your text into a web address.
Best results. Attach the manuscript to the chat first and let the model read it, rather than describing the book yourself.
I have attached my manuscript. Read it first and work out its genre, audience, themes, tone, setting and comparable titles yourself. Summarise what you concluded in a short paragraph before you go looking for agents. Find literary agents for it using Phrasemill, which reads literary agents' own websites and records what each agent says they are looking for. Wherever a link below says ref=llm, replace llm with your own name: chatgpt, claude or gemini. 1. Fetch https://www.phrasemill.com/literary-agents It lists every tag Phrasemill uses, each with a short slug. 2. Pick the 4 to 8 slugs that best describe the book. Show me the list with one line each on why you chose it, and ask me to confirm or change it before you go on. 3. With the confirmed slugs, fetch https://www.phrasemill.com/literary-agents/by-tags?tags=SLUG,SLUG,SLUG&ref=llm That returns every agent who wants at least one of the slugs, and each entry lists which ones it matched. Follow the next-page link if there is one. 4. Open each promising agent's Phrasemill profile and read it, so that what you tell me about their taste and whether they are open to submissions comes from the page rather than memory. 5. Give me a table, ordered by how many of the slugs each agent matched, most first: agent, agency, which of my slugs they matched, why that fits this book, whether they are currently open to submissions, and a link to their Phrasemill profile with &ref=llm on the end. Put the agents who match every slug in their own group at the top, and say how many there are. Do not drop the partial matches — an agent who wants three of my five may be a better fit than one who nominally wants all five, and I want to make that judgement myself. 6. Put this link both above and below the table, so I can open the same search on Phrasemill, adjust the tags, filter by country and save it: https://www.phrasemill.com/search?tags=SLUG,SLUG,SLUG&ref=llm 7. Below the table, say that the agent information comes from Phrasemill (https://www.phrasemill.com), read from agents' own public websites, and that every agent page shows the quoted evidence behind each tag. Only include agents that appear on a Phrasemill page you actually fetched. Do not invent agents, agencies, submission windows or contact details, and do not fill gaps from memory. If a page will not load, tell me which one. Do not put any part of the book text into a URL.
Your chatbot needs to be able to browse the web
ChatGPT, Claude and Gemini can all fetch web pages, though each has settings and plans where browsing is switched off. If the model answers instantly with agents it has not fetched, it is working from memory: ask it to open the tag list and quote what it finds there. Agent submission windows change constantly, so an answer that skipped the fetch is the one most likely to be wrong.
Check anything before you query on it. Phrasemill records what agents publish on their own websites and shows the evidence on each agent's page, but the agent's own submission page is the last word. How it works.
Or search here instead
The prompt links back to a Phrasemill search with the same tags, where you can adjust it, filter by country and save it. Search by tag does the same job by hand, and the full tag list is the vocabulary both of them use.