Wild Birds Unlimited · AI field guide

you already use AI every week. now use it well, and safely.

Everything from the session in one place: how it works, what's safe to paste, the prompts, all five live builds and every practice file. Bookmark it for the next time you open ChatGPT or Copilot.

jump to the builds get the practice files

01 · where you landed

you're not starting from zero.

All 42 of you took the survey before the session. Most of you already use AI. The work now is using it well, and knowing exactly what's safe.

81%use AI at least once a week

3.2/5average confidence in your own AI use

3AI tools used, on average, per person

3.0/5clarity on what's safe to put in. this is the gap.

Safety and better prompts topped your list of what you wanted to learn, so that's what most of this guide is built around.

02 · how it works

prediction, not search.

About 3 in 4 of you described AI as "searching the internet." It isn't. Tools like ChatGPT and Copilot are large language models. They learned patterns from a huge amount of text, and when you ask something, they predict the most likely next words, one chunk at a time.

why it sounds so sure

It's built to produce fluent, confident text. Fluent and confident are not the same as correct, and it has no built-in sense of when it's guessing.

why it makes things up

When the pattern points somewhere that isn't true, it still writes it beautifully. That's a hallucination. Fewer than half of you knew the word, which is exactly why it matters.

how to treat what it gives you

Like a sharp new hire's first draft. Fast, often great, never final. You check it, shape it and own it before it goes to an owner.

03 · good at / double-check

confident is not correct.

Most of you already rework what AI gives you instead of pasting it as-is. That instinct is exactly right. Here's where to lean on it and where to slow down.

lean on it for

a rough first draft of almost anything
summarizing long documents, reports and call notes
generating lots of ideas and options fast
matching a voice, once you show it examples
rewriting dense material in plain English

double-check every time

math and totals (38% of you thought it was reliable)
sources and citations, which it can invent
"remembering" past chats, which it does imperfectly
current facts, prices and policies past its cutoff
its own mistakes, which it rarely flags

04 · safety

green means go. red means ask first.

When a store owner's customer list lands in your inbox, 1 in 3 of you weren't sure what's safe. WBU already has a clear line. Here it is in plain English.

before you paste anything, ask these 3 questions

Is it on WBU's red list?
If yes, stop. It needs approval first.

Does it name a customer, employee or store owner?
If yes, anonymize first, then prompt.

Is this tool approved for this task?
If you're unsure, ask before you use it.

When in doubt, don't paste. Rewrite it first.

green: go for it

review responses
event and promo content
operational efficiency
drafts, research and brainstorming

red: never without approval

financials and strategy
customer or employee personal data
ops manuals, trade secrets and board info
AI images of birds or products

Always: say when you used AI, verify every fact, and remember that AI-written text can't be copyrighted.

not every tool treats your data the same

tool	what happens to what you type	protection	verdict
Free or personal ChatGPT	May be used to train future models unless you opt out	Low. Assume it's stored.	never for WBU data
WBU-approved ChatGPT account	Business plans don't train on your data by default	Medium. Better, but follow the policy.	green-list work
Microsoft Copilot, signed in with your WBU account	Stays inside WBU's Microsoft tenant	High. Your safest default.	safest, still no red list

Rule of thumb: if WBU hasn't approved it, treat it like a public forum.

in the gray zone, have a default move ready

ask before you act

Check WBU's AI policy or submit a Support Request Form for approval. Thirty seconds now saves a lot of headache later.

anonymize, then prompt

Swap customer names, owner names, store numbers and dollar figures for placeholders. "Store A" gets you the same output without the exposure.

use the approved tool

For anything sensitive, stick to WBU-approved tools signed in with your work account, never a personal or free one.

do it without AI

Some things shouldn't go through an AI tool yet. Knowing when not to use it is as valuable as knowing how.

recording a call? get everyone's yes first.

Any AI notetaker, including Copilot in Teams, records the room. In about a dozen states everyone on the call has to agree. Confirm the notetaker is approved, add a line to the calendar invite, and open with: "I've got an AI notetaker running. Is everyone okay with that?"

05 · your line

own the thinking. delegate the typing.

you keep

the relationship with every owner
the naturalist expertise
the hard coaching conversations
the final "is this true and on-brand?" call

hand to AI

first drafts and 20 subject line options
turning call notes into action items
one article reworked into three formats
grammar, tone and plain-language checks

If a wrong answer costs an owner's trust or the brand, you own it. If it's volume or format, hand it off.

06 · prompting

a simple structure, then let AI write the prompt.

Almost half of you keep refining until it's right. That works. Giving context up front works faster.

act

Tell it what role to play. "Act as a franchise communications expert who knows backyard bird feeding retail."

context

Who's the audience, what's the goal, what does it need to know. Just keep red-list data out.

think deeply

Ask it to think deeply about the task. When you tell it to think harder, the quality goes up noticeably.

question

What's the first step you need it to take? Simple, direct and concise.

prompts to copy

the meta-prompt

For blank-page days. Describe the job and let AI write the prompt for you.

I need to [describe the task, who it's for and what good looks like]. Ask me any questions you need first. Then write me a detailed, reusable prompt I can save and run every time I do this. Use plain language and leave [BRACKETS] for the details that change each time.

a hard owner reply

When you know what you want to say but can't find the words.

Act as a franchise support specialist for a nature retail brand. Write a warm, accountable reply to a store owner who is frustrated that [their loyalty points report was inaccurate for the second month]. Acknowledge the frustration, explain the next step clearly, offer a quick call to walk through it, and keep it under 150 words. Plain language, no jargon, no over-apologizing.

a social post that sounds like us

Curious and warm, not a textbook.

coaching call prep

The hard conversation, a little easier.

Act as an experienced franchise business coach. Help me prepare for a call with a store owner whose sales are down year over year and who tends to get defensive. I want to lead with empathy, stay factual, agree on two or three specific actions, and end with a clear follow-up date. Give me an opening, 3 talking points, and a closing question that invites their ideas.

teach it your voice

Do this once per project, then save the result as its instructions.

Here are [3 newsletters / emails / posts] I'm proud of. Describe the voice in 5 bullets: tone, sentence length, words we use, words we avoid, and how we close. Then write those bullets as standing instructions I can paste into a project so everything you write for me sounds like this.

07 · five live builds

built together. yours to rerun.

Each build works on your own green-list work or on our practice files below. Run the 3 questions first, then save what you built so it runs again next week.

coaching + FranDev

call notes to follow-up

paste in a call transcriptYours or our mock call. 3 questions first.
ask for the recap and action itemsWho owns what, by when, in a simple table.
turn it into the next stepFollow-up email, broker update or onboarding checklist.

rerun prompt

Act as a franchise operations assistant. Below is a call transcript. Give me a 5-line recap, then a table of action items with owner, task and due date. Flag anything that was promised without a date. Then draft [a follow-up email / a broker update / an onboarding checklist] in a warm, plain voice, under 150 words. [paste transcript]

save it as a project so every recap starts here next time.

practice files

ops, accounting + marketing

five reports into one

drop in your exportsGreen-list reports or our mock exports.
ask what changed and what looks offTrends, outliers, anything that doesn't add up.
draft the store emails from itOne template, filled per store, checked by you.

rerun prompt

Act as a retail operations analyst. Think deeply. I've uploaded [five] monthly reports for our stores. Tell me: 1) what changed versus last month and last year, 2) any store that looks off and why, 3) any numbers that don't add up, are missing or look like typos. Show your math and list anything you're unsure about. Then write one short email template I can fill in for each flagged store.

check the math, then save the prompt to rerun every month.

practice files · some have mistakes planted in them

comms, training + nature ed

one source, new pieces

pick your trackA: a public bird topic from the website. B: the practice handbook page.
build the new piecesA: Facebook Live outline, newsletter, FAQ. B: a plain-English rewrite.
fact-check it and make it sound like youVerify every claim. B: flag discontinued products.

track A prompt

Act as a naturalist and content writer for a backyard bird feeding brand. Using only the facts I paste below, create: a 5-point outline for a Facebook Live, a 150-word newsletter blurb, and 5 FAQ questions with answers. Mark any claim I should verify before it goes out. [paste public article]

track B prompt

Rewrite this handbook page in plain English for a new store employee. Keep every rule, cut the jargon, use short steps. Then compare it to the discontinued products list I've attached and flag every product it mentions that's being discontinued. [paste handbook page]

save your voice in a project so every piece sounds like you.

practice files · track B

set it up once, use it every week

your first custom GPT

pick one repeat taskThe email, recap or report you do every week.
write the instructionsRole, steps, tone and what good looks like.
add safe reference filesPublic content and templates. No manuals or financials.
test it on fake examplesRun it five times and fix what it gets wrong.
share it with your teamSame quality, whoever runs it.

The full starter instructions are in the custom GPT section below.

share it with your team so everyone gets the same result.

practice files

laptops open

finish yours

Pick the build that fits your week, swap in your own work, and leave with it running. Touches red-list data? Get it approved first.

my one thing

the play I'm running this week:

play

by friday I will have

I'll know it's working when

If it's working, add a second. If it's not, ask ChatGPT why and adjust. That's the loop.

08 · practice kit

every practice file, in one place.

Every store, owner, customer and number in these files is made up, so they're safe to paste into your WBU-approved ChatGPT or Copilot. Real owner financials, customer lists and ops manuals stay on the red list.

download the whole kit

All 11 files in one zip: two call transcripts, five reports, the handbook page, the discontinued list, the GPT starter and five test emails.

09 · custom GPT starter

an owner support GPT, ready to adjust.

In ChatGPT, go to GPTs, then Create. Paste this into the Instructions box, then change anything that doesn't sound like you. Test it on all five sample emails before you share it.

starter instructions

Owner support replies in WBU's voice.

You help the WBU owner support team reply to emails from franchise store owners. Before drafting, ask me for anything you need: the store, the issue, and what we can actually promise. Never invent policies, dates, fixes or phone numbers. If you don't know, leave a [BRACKET] for me to fill in. Tone: warm, plain language, accountable, never defensive or robotic. Sound like a person who knows birds and retail. Structure: acknowledge the issue, say what happens next, offer a clear next step. Keep it under 150 words. If the email involves customer personal data, financials or anything on WBU's red list, stop and remind me to check the policy first. Give me one draft, then ask if I want it shorter, warmer or more direct.

GPT or Copilot agent: which one, when

	custom GPT	Copilot agent
lives in	ChatGPT	Outlook, Teams and your files
best for	Repeat writing and content	Work that lives in your inbox and files
think of it as	A recipe card	A teammate in Outlook
needs	A paid ChatGPT plan	Microsoft 365 Copilot

10 · toolkit

the tools already on your desk.

ChatGPT is the one 90% of you have tried, and Copilot is already inside WBU's Microsoft account. Sign in to both with your work login.

ChatGPT (WBU-approved account)

Your everyday partner for drafts, content, summaries and custom GPTs.

Tip: one project per job, so it always knows the basics. Never use a personal or free account for WBU work.

Microsoft Copilot

Built into Outlook, Teams, Word and Excel, and it stays inside WBU's Microsoft tenant.

Tip: your safest default for anything in your inbox or files. Red-list data still needs approval.

projects

A workspace that keeps files, standing instructions and past chats together, so you stop re-explaining.

Tip: start with owner support, newsletter or coaching prep. Approved files only.

deep research and Researcher

AI runs dozens of searches and hands back a full report. ChatGPT calls it deep research, Copilot calls it Researcher.

Tip: click the sources before you repeat anything.

voice typing

You talk at about 150 words a minute and type about 40. It's already on your laptop.

Tip: Windows: Win + H. Mac: Edit menu, Start Dictation. ChatGPT: tap the mic.

AI notetakers

Copilot in Teams can recap meetings and pull out action items.

Tip: confirm it's approved and get everyone's yes before you record.

11 · agents, safely

last year AI answered. this year it does.

An agent takes a goal and works through the steps on its own: reads, decides, drafts, hands off. You can build one with no code using custom GPTs, ChatGPT's agent mode or Copilot agents.

anatomy of an agent: an owner support inbox

triggerAn owner emails support.
extractPull out the store and the issue.
decideHow-to, bug, or escalate?
actDraft a reply, tag it, route it. A person approves.

safe to start

meeting and call recaps
draft replies that you send
research scans on public topics
sorting and tagging your inbox

keep a person on it

emailing owners or customers on its own
anything that spends money
posting as you or as the brand
touching red-list data or live systems

watch for hidden instructions

Web pages and emails can hide text that tries to steer an agent. Limit what it can open and connect only tools WBU has approved.

humans paper over the cracks. AI won't.

An agent needs a process that's consistent and written down. Point it at a messy one and it surfaces every gap you've learned to live with.

12 · glossary

words you can explain to a coworker.

LLM (large language model): The engine behind ChatGPT and Copilot. It learned patterns from a massive amount of text, so it predicts what should come next.autocomplete, trained on a library the size of a city.
training data: Everything the AI read before you typed a word. It has a cutoff date, and free tools may keep learning from what you type.what a new hire read before day one.
prompt: Whatever you type, say or upload to tell AI what you want.ordering at a restaurant. Specifics get you what you wanted.
token: The small chunks AI reads and writes, usually part of a word. About 100 tokens is 75 words.LEGO bricks. You see the castle, AI sees the bricks.
context window: How much AI can hold in its head at once: your messages, its replies and any files you shared.a desk. When it gets crowded, older stuff slides off.
hallucination: When AI states something confident and well-written that is completely made up.a friend who hates saying "I don't know."
model: A specific version of the AI's brain. Some are fast and light, some slower and smarter.a company fleet. You don't take the semi to grab coffee.
thinking mode: A setting that works through a problem step by step before answering. Slower, better on hard questions.scratch paper instead of mental math.
project: A workspace that keeps related chats, files and standing instructions together.a labeled binder for one job.
memory: Lets AI remember things about you across chats. You can see, edit or turn it off.a barista who knows your usual.
web search: Lets AI look things up on the internet right now and show you its sources.pulling up today's menu instead of guessing.
deep research: AI runs dozens of searches, reads lots of sources and hands back a full report.hiring the librarian for the afternoon.
connector: A secure link between AI and an app you use, like Outlook or SharePoint.a key to one filing cabinet, not the whole building.
agent: AI that takes a goal and works through the steps on its own.a teammate you train once, who checks with you before anything goes out.

13 · your monday

monday morning, rewritten.

Pick two or three tasks you do every week and get really good at those. The rest can wait.

Set up one ChatGPT project for the job you do most.
Teach it your voice with 3 to 5 samples you're proud of.
Run the 3 questions before you paste anything.
Use the meta-prompt on blank-page days.
Rerun your build 05 at least once this week.

Team norm worth trying: label AI-assisted work with "AI-assisted, reviewed by me."

be smart. not bored.

Train In Your Lane for Wild Birds Unlimited · practical AI training that helps your team move forward with confidence.