What agencies and consulting firms use it for
The work that fits has a known shape and a client-specific input: a proposal built from past proposals, a report built from this month’s exports, copy built from a brand guide. Each job below lists the templates and tools that do it.
Proposals and RFP answers
Most RFP questions have been answered before, in a past proposal someone has to find. Paste your approved answers and case summaries, and the template drafts each response from them alone, marking the questions it cannot answer from approved content. Before the pitch, a prep brief built from the prospect’s own site and filings tells the team what the client is likely to care about. Quotes and estimates go out with the assumptions written down.
Draft RFP answers from approved contentPrepare for a discovery callSend a quote that sets expectations
Research and synthesis
Twelve stakeholder interviews, a competitor’s pricing pages and 600 survey comments are the raw material of a strategy deck. Synthesis prompts group what people said, count it, and quote it, so the partner reviewing the deck can trace every finding to a source. A claims check against the pasted sources catches the line in the draft that no interview actually supports.
Synthesize interviews into findingsCompare competitors from pages you pasteFind themes in survey responsesCheck claims against sources
Monthly client reporting
The report a client reads is the paragraph that explains why cost per lead rose 18% and what you are doing about it, not the dashboard. Paste this month’s and last month’s exports with the goals from the statement of work, and the prompt drafts the explanation, the executive summary and the status update. The numbers come from the export; the account lead adds the why that only they know.
Explain why a metric movedSummarize for a busy executiveWrite a status update from notes
Copy at volume, in each client’s voice
Ad variants inside each platform’s character limits, landing pages, SEO titles and line edits that keep a client’s voice all come from one habit: a fixed voice block per client (tone, banned words, claims legal has approved) at the top of every prompt. That block is identical on every call, so prompt caching makes it cheap to repeat. Compare two or three models blind in Model Compare before you standardize on one for a client.
Write ad variants within character limitsDraft landing page copyEdit without losing the voiceModel Compare
Meetings and follow-ups
Build the agenda from what the meeting must decide, then turn the notes or transcript into decisions, owners and dates, and send the client a recap the same afternoon. The recap is where scope disputes are prevented, so it should state what was agreed in the client’s own terms.
Build an agenda from what it must decideTurn notes into decisions and action itemsSend a recap email after a call
Scope, roles and contract terms
When a client asks for “one more round” on a fixed-fee project, a prompt that compares the request with the signed scope gives the account lead a clear answer and a polite email (the sample prompt below). A RACI chart from the kickoff notes settles who approves what. Before legal sees a client’s paper, check its terms against your standard positions so the call with counsel is short.
Turn project notes into a RACI chartSay no and keep the relationshipCheck a contract against your standard terms
Case studies and the agency’s own marketing
Turn a client interview into a case study that uses their words and their numbers, with their written approval, and turn a finished project into a LinkedIn post that says what was hard about it. One article can then become posts for several channels without saying anything new.
Turn a customer interview into a case studyWrite a LinkedIn post about a projectTurn one article into posts
A worked example: the monthly paid-media report
Take a hypothetical 30-person performance agency with 22 retainer clients. Every account lead spends most of a day each month writing a report from platform exports, and the quality varies by who wrote it. The agency wants one tested prompt that drafts the narrative from the exports, leaving the lead an hour of editing instead of a day of writing.
- 1Start from the metric explainerPaste the client’s exports for this month and last, plus the goals in the statement of work.
- 2Grade itThe free estimate checks that the prompt names the reader, the length and what to do when a number is missing. No key needed.
- 3Test it on six past monthsEach month is one case. Checks: no promises, under 400 words, and a judge for numbers missing from the export.
- 4Let the judge find the best valueRun one month on three models. Best value is the cheapest answer within a point of the top judge score.
- 5Price it per clientWork out the monthly cost per account so it sits inside the retainer.
- 6Keep a version per clientSave each client’s variant with its history, and diff it against the agency standard.
The test that earns its keep is faithfulness: an AI judge criterion such as “every number in the output appears in the export” flags the report that rounds a 2.4% conversion rate to 3%. Add a “Does not contain” check for words your agency never puts in a client report, such as “guarantee.” Read the judge’s reasons on the cases it fails; judges are useful and fallible. The Prompt Tester and Model Compare run on your own API key, billed by your provider.
What it costs
The chart prices a hypothetical 30-person agency drafting about 150 documents a working day across all clients (reports, proposals, recaps, copy batches and synthesis) for 22 working days. Inputs are long, because exports, transcripts and briefs are long: about 6,000 tokens in and 900 out per call. The client voice block and instructions are 30% of the input and can be read from cache. That is about 20 million input tokens and 3 million output tokens a month.
A hypothetical agency: 150 drafts a working day for 22 working days. Monthly cost of 3,300 calls with about 6,000 tokens in and 900 out each, 30% of the input read from cache. List prices checked October 20, 2026; your bill may differ.
Set that against the hours the drafts replace and the model cost is rarely the deciding number. The deciding questions are whether the model passes your faithfulness tests and whether each client’s contract allows their data in it. Use the cost calculator to price a single client’s workload.
A sample prompt
A scope check for the moment a client asks for something extra. The double-brace words are blanks you fill in.
You help an account lead at {{agency_name}} respond when a client
asks for work that may be outside the signed scope.
Signed scope (the statement of work, verbatim):
"""
{{scope}}
"""
What the client asked for:
"""
{{request}}
"""
Write three parts:
1. One sentence: in scope, out of scope, or unclear, quoting the
scope line that decides it.
2. If out of scope or unclear: a short, friendly email to
{{client_contact}} that says what is covered, what is not, and
offers these next steps: {{options}}. Mention no price unless
it appears in the options.
3. A note to the account lead listing anything you were unsure
about.
Do not promise dates, discounts or extra work. Under 200 words in
total.
Quoting the deciding scope line is what makes the answer usable: the lead checks one sentence against the contract instead of rereading it.
Guardrails
Practical guidance, not legal advice. Your client contracts decide most of this.
Client data follows the contract
Master service agreements and NDAs often say where client data may go, and some clients restrict AI use on their account entirely. Check before pasting a client’s customer list, unreleased results or campaign data into any tool, use agency accounts rather than personal ones, and strip what the task does not need. The thinkhard.si tools send prompts from your browser to the provider on your key and never to thinkhard.si, but the provider still receives them.
Endorsements and influencers need disclosure
Agencies that recruit and brief influencers share responsibility for disclosure. The FTC’s Endorsement Guides FAQ says agencies need programs to train and monitor the influencers they pay. If a prompt drafts influencer briefs or captions, the disclosure requirement belongs in it as a fixed rule.
No invented reviews or testimonials
The FTC’s rule on consumer reviews and testimonials bans fake ones, including AI-generated reviews and testimonials from people who do not exist. Case studies and quotes come from real clients who approved them.
Ownership of AI-assisted work
The US Copyright Office’s report on copyrightability concludes that purely AI-generated material is not protected, and that prompts alone do not make the user the author; human contributions are assessed case by case. If your contracts promise clients full ownership of deliverables, talk to counsel about how AI-assisted work fits that promise, and disclose AI use where a contract requires it.
Numbers come from the source
A report with one wrong number loses the client’s trust in the rest. Keep figures tied to pasted exports, test for faithfulness, and have the account lead check every number before it goes out. Invented figures, or hallucinations, are most likely when the prompt asks for a comparison the data does not contain.
Key takeaways
- The best fits have a known shape and a client-specific input: proposals, reports, recaps, copy batches.
- Keep a fixed voice block per client at the top of the prompt; it keeps copy consistent and caches cheaply.
- Test reports for faithfulness to the export before any client sees one.
- Client contracts decide where client data may go. Check them before you paste.
- Never generate testimonials or reviews, and build disclosure into influencer briefs.
Related templates, guides and posts
Questions
Can we use client data in AI prompts?
Only where the client’s contract allows it, in tools your agency has approved, and only the data the task needs. Some clients restrict AI use on their account, so check the MSA and any NDA first. Anonymized or aggregated data is enough for most drafting.
Who owns deliverables made with AI help?
The US Copyright Office has concluded that purely AI-generated material is not protected by copyright and that human contributions are judged case by case. How that interacts with the ownership terms in your client contracts is a question for counsel; this page is not legal advice.
How do we keep each client’s voice consistent across the team?
Write a fixed voice block for each client and put it at the top of every prompt. Keep the master copy in your shared drive, save working versions in My Prompts with their history, and send colleagues a share link or an export. Compare versions in the prompt diff before anyone changes a client’s block, and test changes in the Prompt Tester on a few past pieces of that client’s copy.