Event-based lead research: find who's in town, and who to meet

What this does

An AI agent (or an AI chat tool with web search) looks up upcoming events in your area, researches the companies that are publicly listed at them, and finds the role, and sometimes the named person, you would want to meet at each. It drafts a short note for each one. You read the list, check it, and reach out yourself.

The idea: when a conference or trade show is in town, your customers and prospects may be too. A short, honest note that mentions the event is easier to write and easier to answer than a cold one.

Who it’s for

• SDRs, AEs, sales managers, and owners who sell into a region (the examples use Chicagoland) and want to know which events to care about and who to meet at them.
• Anyone selling to businesses that show up at local trade shows, conferences, and association meetings.

Not for you if you want a tool to connect with people and message them on its own, or to build a big list and blast it. This play stops at a review list on purpose. See “Why not full auto?” below.

What you need

• Your area and date range (for example, “Chicago and the suburbs, the next 6 weeks”)
• Who you sell to and what you sell, in one sentence
• The role you want to meet at each company (for example, “Operations Director” or “Owner”)
• Your do-not-contact list, if you have one (customers who asked to stop, companies on a hold list)
• Tools:
  • Agent version (primary): Grok Bot, an AI agent that can run a saved task on a schedule. xAI’s docs say a Bot works on a cloud computer with a browser, keeps working when your laptop is closed, and can run a saved workflow as a routine on a schedule.
  • Manual version: any AI chat tool with web search (ChatGPT, Claude, Gemini, or Grok). You run the prompts yourself, each time you want a fresh list. The scheduled version needs an agent that can run a scheduled task, like Grok Bot. A plain chat tool can’t run itself next Monday.
  • A spreadsheet or doc to hold the review list.
• Cost: see the front matter. Free tiers and plans change; check before you rely on them.
• Time: see the front matter. All times are estimates until we time a real run.

Steps

These steps cover the manual version first. The agent version in Steps 8 to 10 is the same job on a schedule.

1. Fill in your variables. Write down your area, date range, who you sell to, what you sell, the role you want, and your do-not-contact list. Keep it in a doc so you can reuse it.
2. Event search. Open an AI chat tool with web search turned on. Paste Prompt 1. You get a short table of upcoming events with an official link for each.
3. Check the events yourself. Open each official link. Confirm the name, dates, and venue. Delete any event you can’t confirm. Pick the 2 or 3 that best match who you sell to.
4. Company research. Paste Prompt 2 with the public exhibitor, sponsor, or speaker names from the event pages you picked. You get a short, sourced note on each company.
5. Find the right person. Paste Prompt 3. It finds the role to look for at each company and a named person only when a public page (not LinkedIn) names them with a title. You get a source link for every name.
6. Check against your do-not-contact list. Remove anyone on it. Remove any company where a current customer contact asked you to stop.
7. Draft the note. For each person you decide to contact, paste Prompt 4. You get a connection note, an optional email, and a one-line “let’s meet at the event” opener. Edit until it sounds like you. Delete anything you didn’t read or don’t mean.
8. (Agent version) Do one run by hand first. In Grok Bot, give your Bot the job described in Prompt 5 as a one-time task. Read the result. Correct it until the review list is something you would use. xAI’s docs say to start with a one-time task and make it reliable before you automate it.
9. (Agent version) Save it and schedule it. Ask the Bot to save the process as a skill, test it on a second set of inputs, then create a routine, using Prompt 5. The docs say a routine runs on a schedule in the time zone set in your Bot settings, the Bot shows you its next run, and a new routine waits for its next scheduled time rather than running right away. Use the Test button to try it now. A test does real work, so use safe inputs.
10. (Agent version) Review weekly. The routine posts a review list in your chat. You open it, check it, and do Steps 3 to 7 on the entries you want. Pause or delete the routine if you stop using it.
11. Reach out by hand. For LinkedIn, search for the person yourself, confirm it’s the same person and company, and send the note from your own account, a few a day at most. Other options are under “Alternatives to LinkedIn” in Watch-outs.

The prompt

There are five prompts. Run them in order. Each one does one job. Replace the [VARIABLES] in square brackets.

Prompt 1: Event search

Use an AI tool with web search turned on.

You are a research assistant for a salesperson. Use web search. Use only public web pages. Do not log in to anything. Do not open or search LinkedIn.

Find upcoming business events and conferences in [AREA, e.g. Chicago and the suburbs] between [START DATE] and [END DATE] where people who buy what I sell might attend.

Who I sell to: [INDUSTRIES AND BUYER TYPES]
What I sell, in one sentence: [PRODUCT]

Rules:
- Only list an event if you found an official page for it (the organizer's or venue's own page) that shows the dates. If you can't confirm the dates on an official page, leave the event out.
- Do not invent events, dates, venues, or attendance numbers. If something isn't on the page, write "Unknown".
- Don't include private or invite-only gatherings, or events that need a login to see.
- Up to [NUMBER, e.g. 10] events, soonest first.

Output: a table with these columns:
Event name | Dates | Venue and city | Type (trade show, conference, association meeting, other) | Who it says it is for (quote or paraphrase the page) | Official URL | Public exhibitor, sponsor, or speaker list URL (or "None found") | Fit with who I sell to (High / Medium / Low, with one reason)

After the table, add a short "Could not confirm" list of events you saw mentioned but could not verify on an official page.

Prompt 2: Company research

Paste in the names from the public exhibitor, sponsor, or speaker list for the events you chose. Don’t paste anything from a registered-attendee list.

You are a research assistant for a salesperson. Use web search, public company websites, and the public event pages only. Do not log in to anything. Do not open or search LinkedIn.

Event: [EVENT NAME, DATES, OFFICIAL URL]
Companies listed publicly for this event (as exhibitor, sponsor, or speaker): [COMPANY NAMES, ONE PER LINE]
Who I sell to: [INDUSTRIES AND BUYER TYPES]
What I sell, in one sentence: [PRODUCT]
My do-not-contact list: [COMPANY NAMES, or "none"]

For each company, up to [NUMBER, e.g. 10] companies, return:
1. Company name and its website
2. What it does, in one sentence, taken from its own website
3. Where it is based, and its size ONLY if its own site or a public news source states it ("Unknown" otherwise)
4. How it is listed at the event (exhibitor, sponsor, speaker) with the page URL
5. Why it might need what I sell: label this "Hypothesis, not fact" and tie it to a line you actually read
6. "On my do-not-contact list?" Yes or No
7. Unknown or verify: what I would need to confirm

Rules: Don't invent anything. Use "Unknown" when a fact isn't on a page. No funding amounts, headcounts, revenue, or news unless you can give the URL and its date. Skip any company that is on my do-not-contact list and say so. Under 120 words per company.

Prompt 3: Stakeholder finding

You are a research assistant for a salesperson. Use web search and public pages only. Do not log in to anything. Do not open, search, or copy from LinkedIn pages. Do not guess email addresses or name patterns.

For each company below, find the person who most likely owns [PROBLEM AREA, e.g. field service scheduling] and who would be a sensible person for me to meet at the event. The role I want: [ROLE OR TITLE].

Companies and event: [PASTE THE COMPANY LIST FROM PROMPT 2, WITH EVENT NAME]

For each company return:
1. Role to look for (title), and why that role fits, in one line
2. Named person: ONLY if a public page other than LinkedIn (the company's team or leadership page, a press release, or the event's own speaker page) states their name AND title. Give the exact page URL and the words that name them. If no such page exists, write "No named person found" and do not guess.
3. How current the source is (its date, or "undated")
4. A public business contact route, ONLY if the company itself publishes it for business inquiries (a contact page or a published business address). Give the URL. Otherwise "None found".
5. Confidence: High (named on an official page with title) / Medium / Low
6. Unknown or verify

Rules: Never infer a person's name, title, or email from other people or patterns. Never include personal details (home address, personal phone, family, age). Business-relevant public information only. If a person might have left the company, say so. Return a table, one row per company.

Prompt 4: Personalized connection note (and alternatives)

Use this only for people you decided to contact, one at a time. You run it, you edit it, you send it.

Write outreach drafts for a salesperson. Use ONLY the facts I give you. Do not invent shared interests, flattery, mutual connections, or familiarity.

NO-ADDITIONS RULE: Do not add facts, statistics, scarcity or urgency, promises, references or testimonials, or the gender of any person that I did not supply.

Person's role and company: [ROLE, COMPANY]
Event and dates: [EVENT NAME AND DATES]
How the company is publicly listed at the event: [exhibitor / sponsor / speaker, per the event page]
One fact I actually read on a public page: [FACT, or "nothing specific"]
Why I'm reaching out, honestly, in one sentence: [REASON]
What I sell, in one sentence: [PRODUCT]
My name and company: [NAME, COMPANY]
Public business email the company publishes for inquiries, if any: [EMAIL or "none"]

Write:
A. CONNECTION NOTE: 200 characters or fewer, no pitch, no link. Mention the event only if the page says the company is listed there. If I said "nothing specific", don't pretend to know anything about them.
B. EMAIL (write only if I gave a published business email; otherwise write "SKIP - no public business email"): subject under 6 words, honest, no tricks; body under 90 words; one low-pressure ask; and at the end exactly these two placeholder lines: "[YOUR OPT-OUT LINE]" and "[YOUR POSTAL ADDRESS]". Don't write the opt-out line or address yourself.
C. IN-PERSON OPENER: one sentence I could say at the event if we meet. No pitch.
D. CHECK LIST: list any phrase in A, B, or C that sounds like flattery or implies I know more than I do, and anything that needs me to verify first.

Plain words. No emoji. No "I'd love to pick your brain." Don't say "I noticed you're attending" unless the event page lists the company.

Prompt 5: Set up the scheduled routine in Grok Bot (agent version)

First run Prompts 1 to 3 as a one-time task. When the review list is reliable, send this to your Bot, then fill in the brackets.

Save the process we just used as a skill called "Event-based lead research". Include: the event search rules, the company research rules, the stakeholder rules (public non-LinkedIn pages only, no guessing names or emails), the output format, and these standing boundaries:
- Never log in to LinkedIn, open LinkedIn pages, or use any LinkedIn automation.
- Never connect with, message, or email anyone. Never enroll anyone in a sequence.
- Skip anyone on my do-not-contact list: [LIST OR FILE].
- Only use public pages. Give a source URL for every fact.
- If a source is missing or a site blocks you, report that instead of using old or guessed data.

Then create a routine for this Bot: every [DAY, e.g. Monday] at [TIME, e.g. 7:00 AM] in [TIME ZONE, e.g. Central Time], run the skill for [AREA] for the next [NUMBER, e.g. 6] weeks, who I sell to: [INDUSTRIES AND BUYER TYPES], what I sell: [PRODUCT]. Post a review list in this conversation, with the event table, the company notes, and the role or named person for each. Do not contact anyone. Show me the routine's next run time when it is saved.

Sample input (fake data)

All names are fictional. Use something like this to practice before you try real events.

Area: Chicago and the suburbs
Dates: the next 6 weeks
Who I sell to: regional food distributors and cold-storage operators
What I sell: route and delivery-window planning software
Role I want: Director of Transportation or Logistics
Do-not-contact list: Harbor Lane Foods

Sample output

Not run yet by us.

Below is only an illustration of the shape of the output. It was written by hand with fictional example data (fictional events, companies, and people) and is not output from a test run. Nothing here is a real event or company, and any resemblance to a real one is a coincidence.

FICTIONAL EXAMPLE, NOT A REAL RUN

EVENT TABLE
Event name | Dates | Venue and city | Type | Official URL | Public exhibitor list | Fit
Great Lakes Cold Chain Expo (fictional) | Nov 12-13 (fictional dates) | Example Convention Hall, Chicago | Trade show | example.com/expo | example.com/expo/exhibitors | High: page says it is for cold-storage and distribution operators
Midwest Grocery Logistics Forum (fictional) | Nov 19 (fictional date) | Example Conference Center, a Chicago suburb | Conference | example.com/forum | None found | Medium: speakers are listed, exhibitors are not

COULD NOT CONFIRM
- "A food-distribution summit in the fall": seen in a forum post; no official page found, so left out.

COMPANY NOTE (one of the list)
Company: Prairie Meadow Cold Storage (fictional), prairiemeadow.example
What it does: Operates temperature-controlled warehouses (from its About page).
Based in: Unknown. Size: Unknown.
Listed at event: Exhibitor, example.com/expo/exhibitors
Hypothesis, not fact: Its "Services" page mentions "same-day delivery windows," so scheduling deliveries may matter to it.
On do-not-contact list? No
Unknown or verify: Whether it has its own trucks or uses carriers.

STAKEHOLDER
Role to look for: Director of Transportation or Logistics (the Services page talks about delivery windows).
Named person: Pat Sample, Director of Logistics (fictional person). Source: prairiemeadow.example/team ("Pat Sample, Director of Logistics"). Source date: undated.
Public business contact route: prairiemeadow.example/contact (a contact form; no email published).
Confidence: Medium (named on the company's own team page, undated, so confirm they are still there).

DRAFT A: CONNECTION NOTE (about 150 characters)
Hi Pat, Prairie Meadow is listed as an exhibitor at the Great Lakes Cold Chain Expo. I work on delivery-window planning software and would like to connect.

DRAFT B: EMAIL
SKIP - no public business email

DRAFT C: IN-PERSON OPENER
"Hi Pat, I'm [NAME] from [COMPANY]. I saw you're exhibiting at the Expo. What are you hoping to get out of the two days?"

DRAFT D: CHECK LIST
- "I work on" assumes your role touches the product; edit to fit.
- Verify on the Expo page that Prairie Meadow is still listed.
- Confirm on LinkedIn, by hand, that Pat Sample is still at Prairie Meadow before you send.

Setup checklist

• Variables written down (area, dates, who you sell to, product, role you want)
• Do-not-contact list ready, and the agent is told where to find it
• Web search is on for the chat tool, or your Bot can use its browser
• Prompts 1 to 3 run once by hand and the output checked against the official pages
• The routine’s standing boundaries are written in: no LinkedIn login, no contacting anyone, public pages only, source URL for every fact
• (Agent version) Routine created, Time zone checked in Bot settings, next run time confirmed
• (Agent version) Test run done on safe inputs, and the result stopped at the review list
• You have a place to log who you contacted, when, and the answer
• An opt-out line and a postal address are ready for any email
• You have read the Safe Use page

Check before you send

Do these yourself, every time, before anything goes to a real person.

• The event is real and the dates are right. Open the official page. The AI can get dates wrong.
• The company is actually listed. Don’t write “I saw you’re attending” unless a public page lists the company as an exhibitor, sponsor, or speaker. A company on an exhibitor list does not mean this particular person will be there.
• The person is real, current, and the right one. The AI found a name on a public page. Check it’s the same person and still at the company. Titles go out of date.
• Every fact in the note is something you actually read. Delete anything you didn’t.
• Not on a do-not-contact list. Check your own list and your CRM for “asked to stop”.
• Nothing implies a relationship that doesn’t exist. No fake familiarity.
• Email only: your real name and company, an honest subject line, your postal address, and a clear way to opt out (see the FTC guide in Sources). Honor opt-outs promptly. If the person is outside the US, check that country’s rules first.
• LinkedIn only: you type or paste the note in yourself and press send yourself. Keep the pace to a handful a day, and stop if someone says no or doesn’t answer.
• Don’t paste confidential CRM data or customer lists into a consumer AI tool unless your company policy says you can.

How it went

Evidence label: Used in real prospecting (self-reported). This section is the author’s own first-hand report, not a test by us. It is self-reported and not independently measured. We have no run log, dates, or model name for it, and we have not run this play’s manual-send version ourselves, in Grok Bot or in a second tool.

What the author reports. The author uses a version of this workflow in Grok Bot weekly. The author reports:

• Over 100 contacts.
• 50+ LinkedIn connections.
• Multiple replies showing interest.

The author’s version is not this play. In the author’s version, the agent handled the LinkedIn outreach itself, including connecting. This play deliberately does not do that. LinkedIn’s User Agreement (Section 8.2) prohibits bots and unauthorized automated methods to add contacts or send messages, and LinkedIn says members who use automation tools risk having their accounts restricted. See “Why not full auto?” below. In this play the agent never opens LinkedIn and you send every connection request or message yourself, by hand. We don’t know whether the manual-send version would produce similar numbers.

What we are not claiming. We have no conversion, meeting, pipeline, or revenue figures for this workflow. We have no time-saved number. “Multiple replies showing interest” is not a measure of deals or sales. Don’t expect the author’s numbers; the connection count came from a version in which the agent sent the outreach.

Re-check status. Manual-send version: not yet run by us. Second-tool run: not done. Re-check by 2026-12-14 (about 60 days from this update). Until we log a real run of this written version, the label stays “Used in real prospecting (self-reported)”, and the play is not labeled “Tried in an AI tool.”

What to watch for (general cautions, not problems the author reported). These cautions come from the verified material elsewhere on this page. They are expected risks, not observed ones:

• Public-source limits. Exhibitor, sponsor, and speaker lists are often thin or posted late, and many companies don’t publish a named person for the role you want. Check each event’s own terms for whether its lists may be used for outreach.
• List quality. A long list is not the same as a good list. Check every event, company, name, and title against the official page, drop anything you can’t confirm, and remove anyone on your do-not-contact list before you contact anyone.
• LinkedIn ToS risk. Letting an agent connect or message on your behalf can get your account restricted. Send by hand, a few a day at most.

Why not full auto?

Short answer: LinkedIn doesn’t allow it, and it’s easy to get wrong.

Some people set this up so the agent connects and sends direct messages on LinkedIn on its own. We don’t do that and don’t recommend it.

• LinkedIn’s rules. Section 8.2 of LinkedIn’s User Agreement says you agree not to “use bots or other unauthorized automated methods to access the Services, add or download contacts, send or redirect messages,” and not to use software, scripts, robots, or browser plugins to “scrape or copy the Services, including profiles and other data.” LinkedIn’s help page on prohibited software says it does not permit third-party software that automates activity on its site, and that members who use such tools risk having their accounts restricted or shut down.
• Your account is yours. LinkedIn can also limit how many connections you make and how many people you can contact (User Agreement, Section 3.4). An agent that sends in bulk can burn through that quickly.
• Spam and consent. An agent can’t know who agreed to hear from you, who asked you to stop, or which note reads as pushy. A person can.
• Mistakes go out under your name. Invented personalization or a wrong name is worse sent than drafted.
• Grok Bot’s own docs lean the same way. Its Sales Outbound example returns a review list and says “do not send or enroll anyone,” and its guidance says to keep sending behind your approval.

So the AI does the research and drafts the note. You review it, then connect from your own LinkedIn by hand. This is general guidance, not legal advice. Read LinkedIn’s current terms yourself.

Watch-outs

• LinkedIn. No bots, no auto-connect, no auto-message, no scraping, and no letting an agent browse LinkedIn for you. In this play the agent never opens LinkedIn. It finds names on the company’s own site, press releases, and the event’s public speaker pages. You look the person up on LinkedIn yourself.
• Alternatives to LinkedIn. If the company publishes a business email or contact form for inquiries, you can use that. For email, include your real identity, a postal address, and a working opt-out, and honor opt-outs (FTC guide in Sources). You can also skip online outreach and just meet the person at the event: stop by the booth, go to the session, or ask the organizer how to arrange a meeting.
• Privacy. Use only public, business-relevant information: a person’s name, title, and company, from a page that publishes it for business purposes. No home addresses, personal phone numbers, family details, or guessing at personal emails. No scraping from behind a login. Don’t use registered-attendee lists, or any list from the event that its terms don’t let you use for outreach. Respect do-not-contact lists and any “asked to stop” request, now and later.
• Other regions. If contacts are in the EU, UK, or Canada, rules on cold outreach can be stricter. Check before you email.
• Invented facts. AI can fill gaps with guesses. Tell it to write “Unknown” and check every fact against the page.
• Agent access. xAI’s docs say all your Bots share one cloud computer, its files, and its logins. Give the Bot the least access it needs. Don’t leave a LinkedIn session logged in on it.
• Not legal advice. See the Safe Use page.

Troubleshooting

• The event list has an event I can’t find on any official page. Delete it. The prompt tells the AI to leave unconfirmed events out; if it still adds one, tell it “remove anything without an official page”.
• The dates are wrong or the event moved. Trust the official page. Ask the AI to re-check only that event.
• The AI returned a LinkedIn link or says it “found them on LinkedIn”. Don’t use it. Re-run Prompt 3 and repeat the rule: public pages other than LinkedIn only.
• “No named person found” for most companies. That’s the honest answer, not a failure. Use the role to look for, and look the person up yourself or ask for them at the booth.
• The exhibitor list is empty or short. Many events post lists late. Try again closer to the date, or use speakers and sponsors.
• The routine didn’t run. In Grok Bot, check that it isn’t paused, that “When to run” shows the schedule you meant, and that the time zone in Bot settings is right. A new routine waits for its next scheduled time; use Test to run it now (a test does real work).
• The routine reports a site blocked it or a session expired. Per xAI’s docs the Bot hands that step to you. Don’t work around it. Use another public source.
• Usage drops faster than expected. Each routine run spends usage. Run it weekly instead of daily, and narrow the area or date range.
• The notes sound the same. Add one real fact you read on a public page to Prompt 4. If you have no fact, don’t send the note.
• A note feels too long for a connect box. Check the limit in your own LinkedIn account, since limits differ by account type. Cut the product sentence first.

Variations

• Other cities. Swap [AREA]. Add the local venue and chamber of commerce calendars to Prompt 1 as places to check.
• Other industries. Change who you sell to and the role you want. A manufacturing rep might want a Plant Manager; a software rep might want an IT Director.
• Trade shows. Use the public exhibitor list as the company list. Run Prompt 2 on the 10 best-fit exhibitors and Prompt 4’s in-person opener for booth visits.
• Conference speakers. A speaker page names the person and title and gives you something real to mention. Use Prompt 3 with the speaker list.
• Local association meetings and chamber events. Smaller, easier to meet people. Shorter lists.
• Owner version. Run it monthly and pick one event.
• Manager version. Give each rep a different event or region, and have them log outreach in one shared doc.
• Short version. Only Prompt 1 each week. Pick the event, then meet people there.
• Other tools. Prompts 1 to 4 work in any AI chat tool with web search. Only the weekly schedule needs an agent with scheduled tasks, like Grok Bot.

Sources

What we verified, and what each supports. All checked 2026-10-15.

• LinkedIn User Agreement (effective Nov 3, 2025). Section 8.2 “Don’ts” prohibits scraping with bots or scripts and using bots or unauthorized automated methods to add contacts or send messages; Section 3.4 lets LinkedIn limit connections and contact ability and restrict accounts.
• LinkedIn Help: Prohibited software and extensions. States that third-party software that automates activity on LinkedIn is not permitted and that members who use it risk account restriction.
• Grok Bot overview (xAI docs). A Bot works on a persistent cloud computer with a browser, can run while your laptop is closed, and drafts are approved by you before they are sent. A site can block automation or expire a session; the Bot hands those steps to you.
• Grok Bot use cases: Sales Outbound (xAI docs). The starter task returns a review list and says not to send or enroll anyone; a nightly research routine that stops at the review list comes after the output is reliable.
• Grok Bot skills and routines (xAI docs). Skills, routines on a schedule, test runs, pause and edit, and the guidance to automate preparation before execution and require approval for sending.
• Grok Bot routines (Cursor help). Plain-language schedules, the time zone setting, a new routine waits for its next time, and each run spends usage.
• CAN-SPAM Act: A Compliance Guide for Business (FTC). Rules for commercial email, including B2B: accurate sender, honest subject, postal address, opt-out, and honoring opt-outs.

What we could not verify: that Grok Bot’s browser reliably finds and reads local event and exhibitor pages (we haven’t run this play); whether any given venue or organizer allows its exhibitor list to be used for outreach (check each event’s terms); and current plan and usage limits for Grok Bot (see xAI’s and Cursor’s plan pages).

Retest

Re-check by: 2026-12-14 (about 60 days from this update). We have not logged a real run of this written version, so there is no run date. Changelog:

• 2026-10-15: Draft written. Not run yet. LinkedIn step set to review-and-send-by-hand after checking LinkedIn’s User Agreement and help page.
• 2026-10-15: “How it went” updated with the author’s first-hand, self-reported use of a Grok Bot version in which the agent handled LinkedIn outreach (not recommended here). Not independently measured.
• 2026-10-15: Published with the evidence label “Used in real prospecting (self-reported)”. The written version has not been run by us; no “tried in an AI tool” claim is made.

Test log

No runs logged yet.