Briefing packs and research: preparing your principal for a meeting

A briefing pack answers one question for a busy person: what do I need to know to walk into this meeting well? Pulling one together means reading internal papers, looking up the other organisation, checking recent news and cutting it all down to a page or two. AI speeds up every part of that. It also produces confident facts with no source, links that don't say what it claims, and out-of-date information presented as current. The skill is using it for the legwork and checking the result as if your principal's reputation depends on it, because it does.

This guide follows Maya Roberts, Office Manager at Fernway Group (our fictional practice company), preparing a pack for Erin Vaughan, the Managing Director.

What goes in a briefing pack

Before any prompt, decide the shape. A pack that works usually has:

• Purpose. Why the meeting is happening and what your principal wants out of it.
• Who's attending, with their roles.
• Background on the organisation: what it does, its size, recent public news that matters here.
• The relationship so far: history, open issues, anything promised.
• Key points and suggested questions.
• Sources, so anyone can check where a fact came from.

Download the briefing pack template for this structure, with a sources table and final checks built in. The fixed shape matters more than the tool: it tells the AI what to produce, and tells you what's missing.

Worked example: Erin's renewal meeting

Maya splits the job in two: the internal picture and the public research. They need different tools.

The internal picture

The relationship history lives in Fernway's own files: account notes, the current contract summary, recent emails from Tom Elliott (Sales Lead) about a possible discount. That is confidential commercial information. It belongs only in a tool your organisation has approved for it, such as Copilot signed in with a work account, where your organisation's Microsoft 365 permissions and data terms apply. Maya uses Copilot Chat with her work account and pastes or attaches the three documents.

Using only the three documents I've attached, write a half-page summary of our relationship with this customer for our managing director, who is meeting them on Friday.

Cover: how long they've been a customer and what they buy; the current contract end date; any open issues or complaints; anything we've offered or promised in the renewal so far.

After each point, say which document it came from in brackets. If something isn't in the documents, write "not in the documents" instead of filling the gap. Keep figures exactly as written.

The draft is useful but Maya spots a problem straight away: it says the discount "has been agreed". Tom's email says he wants sign-off on it and Priya asked for the numbers first. It hasn't been agreed. Maya corrects it, because Erin walking in believing a discount is agreed would be a real mistake. It is a typical failure in summaries of internal material: a proposal becomes a fact.

The public research

For the customer's organisation, Maya wants public information: what they do, how big they are, any recent news such as a merger, new sites or a change of leadership. Nothing confidential goes into this step, so a general assistant with web search is fine.

Search the web for public information about [organisation name], a UK [sector] business based in [area]. I'm preparing a briefing for a meeting about a supplies contract renewal.

Give me: what they do, approximate size, and any news from the last 12 months that could affect their buying (mergers, new sites, restructuring, leadership changes).

For every fact give the source link and the date of the source. Prefer the organisation's own website, its published reports and official filings over blogs. If you can't find something, say "not found". Don't guess.

The person

For Nadia Carr, Maya keeps it narrow: her job title, what her role covers and anything she has said publicly in that role, such as a quote in her organisation's news release about procurement. That is relevant and professional.

Maya doesn't ask the tool for a profile of Nadia: no personal social media, family, home location, age or anything unrelated to the meeting. Researching a named individual means handling their personal information, and UK data protection law expects personal data to be limited to what the purpose needs (the ICO calls this data minimisation). If a result drifts into personal territory, Maya leaves it out.

Web search or deep research?

Most chat assistants can now search the web, and several offer a longer "deep research" mode that reads many sources and writes a report.

Web search suits quick, specific questions: a company's head office, a recent announcement, a spelling of a name. It answers in seconds with links. Use it for most briefing packs.

Deep research suits a bigger question where you want many sources weighed: the state of a sector, a comparison of several organisations. It takes minutes, produces a long report and still makes mistakes. Its length makes errors harder to spot, not less likely. For a two-page brief it is usually more than you need.

Neither will usually see paywalled articles, private company information or anything behind a login. Both can surface old pages as if they were current. The lessons on ChatGPT web search and deep research, Claude research mode and Gemini Deep Research cover each tool.

If your research sources are documents you already hold, a tool that answers only from uploaded sources, such as NotebookLM, keeps the answers tied to them.

Checking sources

This is the step that makes a pack trustworthy, and the one it is most tempting to skip.

Tools sometimes give a real link to a page that says something different, or cite a page that doesn't exist. Opening every link sounds slow. For a one-page brief it is ten minutes, and it is the ten minutes that matter. The source that did not exist shows how close a fake citation can get to a senior reader. Spot the Hallucination is good practice for noticing the signs, and the free Prove It course teaches a full checking routine.

Citing where things came from

Put the source next to the fact in the pack itself: "Opened a new distribution site in Coventry in March (company news page, dated 12 March)". Your principal may be asked "where did you hear that?" in the meeting. A source in brackets lets them answer, and lets anyone check the pack later.

Pulling it together

Once the internal summary and public research are checked, Maya combines them into the template and asks for one last pass: "Cut this to one page. Keep every source label. Add three questions Erin could ask about their plans for next year, based only on what's in the pack." She reads the result once more, fixes a date and sends it on Thursday morning.

Data safety

Keep two kinds of material apart. Public research can go in a general assistant. Internal papers, contract terms, pricing and anything about your organisation's plans need an approved tool with a work account. Don't paste internal documents into a personal account to "just get the summary". Your organisation's policy and approved tools still take priority, and if you're unsure, ask before you paste.

Common mistakes

• Trusting unsourced facts. If the tool can't tell you where something came from, treat it as unconfirmed.
• Letting proposals become agreements. Internal summaries often upgrade "suggested" to "agreed". Check the verbs against the source.
• Researching people too widely. Professional role and relevant public statements only.
• Using old news as current. A leadership change from three years ago, presented as recent, can embarrass your principal. Check dates.
• Making the pack too long. A busy principal reads the first page. Put the purpose and the three things they must know at the top.
• Mixing tools carelessly. Internal material in a personal chat assistant is a data problem, however useful the summary.