Using AI to brief yourself on every guest before the dinner

Gartner predicts that by 2027, 95% of sellers' research workflows will begin with AI, up from under 20% in 2024 (Gartner, 2026). That is a forecast rather than a measurement, but the direction is not in much doubt. Sellers in Salesforce's latest sales survey expect AI agents to cut prospect research time by 34% once fully implemented (Salesforce, 2026, vendor research).

For a field marketer, the densest research job of the quarter is the executive dinner. Sixteen guests, a host who needs to know who is across the table, and an account team that wants every conversation to count. Done by hand, the briefing pack eats days. Done carelessly with AI, it produces confident paragraphs with the wrong job titles.

This workflow is for the version in between: AI does the reading and drafting, and a person verifies the facts that would embarrass the host if they were wrong.

Step 1: Build the input sheet

Start from the confirmed guest list, joined to your CRM. One row per guest, with these columns:

• Name and company, as registered
• Title in the CRM, plus the date that field was last updated
• Account owner
• Open opportunity, if any: stage, and the problem in play
• Past attendance at your events
• A note from the account owner: relationship context, anything sensitive, anything not to raise

The last column matters most, and no model can fill it. Send the sheet to each account owner with a single question: what should the host know that is not in the CRM?

Step 2: Gather sources and set the boundary

Collect public material at the company level: the newsroom, recent earnings or funding announcements, product launches. At the person level, use only what the guest has published about their own work, such as talks, articles and posts on professional topics.

Draw the line there and write it down. No family, health, home location, politics or personal social accounts. Data security and privacy is the top AI concern among event planners, at 59% (PCMA, 2025), and a host who mentions something a guest never shared professionally has turned a dinner into surveillance.

If you use an assistant with web browsing, put the same boundary in the prompt and require a URL for every claim. If you paste sources in yourself, which is slower but easier to check, put the URL above each excerpt.

Check your company's rules on which tools may receive CRM data. Only 38% of marketers say their company has generative AI policies (Marketing AI Institute, 2025), so the rule may not exist yet.

Step 3: Draft the brief, one guest at a time

Run each guest separately. Batching the whole list into one prompt raises the odds that one guest's details bleed into another's paragraph.

You are preparing the host of a private dinner on [topic].

Inputs: this guest's CRM row, and source excerpts with URLs.

Write one paragraph, no more than 100 words, on the guest's
company: what it does and what has changed in the last six
months. Then one line on the guest's own published work, if any
appears in the sources.

Rules:
- Use only the inputs. End each sentence with its source URL
  in brackets.
- Do not state or guess the guest's current job title or how
  long they have held it. I will add this after verification.
- Nothing about family, health, politics, location or personal
  life, even if it appears in a source.
- If sources conflict, say so and quote both.
- If a section has no source, write "nothing found".

The job-title rule looks fussy. It is the most important line in the prompt. Models blend training data that can be years old with whatever pages they find, and professional profiles lag reality. The CRM is no safer: 48% of CRM administrators said customer data decayed faster over the prior year (Validity, 2024, vendor research). A host who congratulates someone on a role they left in the spring does not get the evening back.

Step 4: Tie talking points to the account

Next, connect the brief to what your company knows. This step uses internal deal data, so it runs only in an approved tool.

Here is the brief for [guest], plus our account context:
- Open opportunity stage: [stage]
- Problem they are trying to solve, per the account owner: [x]
- Account owner's note: [note]

Write three conversation openers for the host. Each should
connect to something in the brief or to the problem above.
None may mention our product, pricing or the deal.

Then write one line headed "Do not raise" listing any topic
from the account owner's note the host should avoid.

Keep the openers on the guest's problem. Attendees say they value substance: 63% call subject-matter experts the most important element of good networking, and 54% are willing to share on-the-job challenges (Freeman, 2025). The dinner works when a guest describes that challenge, and it stalls when the host steers toward a demo.

Brief the account as well as the person. A typical buying decision now involves 13 internal stakeholders and 9 external influencers (Forrester, 2026). When two guests come from the same account, brief them together and note how their roles differ. Delainey Kirkwood of Gartner's sales practice has made the case for thinking at the group level: "Messages tailored to the buying group or organization foster understanding and consensus, but individual-level content can lead to confirmation bias." (Gartner, 2025)

Step 5: Find the connections across the table

Once every brief is done, run one prompt across the full set.

Below are briefs for all [n] guests. Identify pairs or small
groups who share a problem, an industry shift, or a topic they
have published on. For each, write one sentence the host could
use to introduce them. Use only what is in the briefs.

This is often the most useful output of the whole exercise. A host who can say "you two should compare notes on this" has made the evening worth the guests' time, and has done it without selling anything.

Step 6: Verify, then package

Before anything reaches the host, one person runs this checklist for each guest:

1. Current title and employer, confirmed on the guest's own current professional profile or company page. If sources disagree, leave the title out and ask the account owner.
2. Every sentence in the brief has a working URL, and the page says what the sentence says.
3. Nothing personal survived the rules. If it did, delete it.
4. The account owner has read the "Do not raise" line.
5. Each "nothing found" stays as written rather than being filled in from memory.

Package the result as one page per guest, in seating or arrival order, and send it to the host the day before. Keep it somewhere your CRM access controls apply, not in someone's personal chat history, and delete the working files after the event.

What AI gets wrong here

The failures repeat. Models conflate people who share a name, especially at large companies. They attach a subsidiary's news to the parent. They describe an undated launch from two years ago as recent. And when told to be specific, they produce specificity whether or not a source supports it. That is why the prompts demand a URL per sentence: it turns every claim into something you can check in seconds.

Buyers are wary of the same thing. In a Gartner survey, 51% of B2B buyers said they were likely to encounter misleading information from generative AI (Gartner, 2026). The brief in your host's hand deserves at least the skepticism your guests bring to their own AI research.