How to Create Prompt Personas in Promptwatch
How we create Promptwatch personas so buyer and intern prompts stay on separate rows and never get averaged together.
The same question from a buyer and from an intern is not the same prompt. Models answer like the user they were handed. We create personas in Promptwatch so those rows stay apart. We do not mix them into one average and call it visibility.
We start with two: the person who can buy, and the person who is researching for someone else. On a B2B account that is often "economic buyer" vs "intern / coordinator." On a clinic it is patient vs practice manager. We do not start with twelve. Name two personas from real roles on the account. Buyer vs intern, or the local equivalent. If you want the first two written, send the roles. Do not send a list of twelve.
The reason for two is that a persona is a row multiplier, and every row spends from the response budget. Start with two and you double the rows on the prompts you attach. Start with twelve and you have made the response cap the bottleneck before you have learned anything. Two personas also force a real choice: who on this account matters first? That question is more useful than the question of how many personas you can name. The two we pick come from roles we have heard on the account, not from a template.
How we write a persona
A persona is who the model should treat as the asker. Job, context, and the constraints they would mention. We steal the language from sales calls and from GSC queries after import. We do not write a novel. We do not invent a demographic we have not heard on the account.
Then we attach that persona to a small set of prompts, usually the five that already make revenue conversations. Same stem, two persona rows. Leave the rest on default until we have a week of rows. If the buyer row cites you and the intern row cites a Reddit thread, that is a ticket. Averaging them into 50% would hide it. The 50% is the number that looks reasonable on a slide and tells you nothing about which half of your audience you are losing.
The five-prompt scope is deliberate. You want the persona attached where the answer changes meaningfully when the asker changes, which is the revenue conversation, not the brand query. A prompt like "what is the best tool for X" gets a different answer from a buyer who names budget and a different answer from an intern who names a free tier. A prompt like "who makes X" barely moves. Attach personas to the prompts where the asker's role would change the model's answer, and leave the rest on default until the rows tell you otherwise.
Location tracking is a different axis. Country-level targeting starts on Essential. State and city targeting start on Professional for brand plans and are also included on the agency plans. Do not stack a new city and a new persona on day one or the response budget disappears. Local pack work stays on the local SEO playbook. A persona is not a city. The two axes interact on the response cap, which is the practical reason to add them one at a time.
GSC import can suggest topics. It will not invent personas. We still type those. Someone with write access creates them in the UI. Analysts who only chat over MCP get a read-only key and cannot add personas from a laptop. That is intentional. See Search Console topics with MCP if the analyst wants to ask Claude about the split later.
The split between write access and read-only MCP access is a guardrail, not a nuisance. Personas are editorial decisions. They encode a judgment about who the asker is, and that judgment should be made by someone who has heard the account's calls, not by an analyst pulling a list at speed. Read-only MCP lets the analyst ask questions about the existing split without being able to change it, which keeps the persona set stable while the analysis runs.
Can we reuse one persona across every client on Kick-off? You can copy the idea. You should not copy the text. A dental practice manager is not a SaaS intern. We write personas from that account's calls.
We will not create an "SEO specialist" persona so we can watch ourselves. Useless on a client scoreboard. We also will not average persona rows in Looker. If the report cannot show two lines, we show a table. Read mention vs citation per persona. Do not average. And we will not attach every imported GSC query to every persona. That is how Kick-off still blows the response cap.
File one content ticket if the intern row cites a source the buyer never sees. Passage work: how to get cited by ChatGPT. Google's AI optimization guidance still applies to the page you edit after that ticket. Add a third persona only when sales can name the role without thinking.
Essential at $95/mo is enough to learn the habit on one country-level project with 50 prompts. Professional is $245/mo when you want more prompts, state or city targeting, or Data Studio. Business is $579/mo. Kick-off is $199/mo with state and city targeting, Data Studio, unlimited projects and prompts, and 10 seats. Explore is free, 10 ChatGPT prompts. Two personas will eat that free allotment immediately. G2: 4.7/5.
Paid engines stay the usual set. Personas do not add engines. They multiply rows.
FAQ
Can we reuse one persona across every Kick-off client?
You can copy the idea. You should not copy the text. A dental practice manager is not a SaaS intern. We write personas from that account's calls.
Do personas add engines?
No. Paid engines stay the usual set. Personas multiply rows.
Should we start with twelve personas?
No. We start with two: the person who can buy, and the person who is researching for someone else. Two personas will eat the free Explore allotment immediately.
If you want the first two written, hello@1001seomedia.com. Send the roles. Do not send a list of twelve.