AI, Ageism, and the New Career Risk: A 90-Day Guide to Staying Valuable

By Robert Carey | November 20, 2026

An experienced professional reviewing information on a laptop.
Illustrative image from the existing Patient Library.

What you'll learn

I watched a younger colleague pick up AI quickly and start producing reports that got attention. They were good reports. He learned a tool, used it well, and looked sharp. I could also see why the person who had spent 20 years learning the business might feel uneasy. If a manager sees a polished report in an hour, does that manager still understand the years of experience it took someone else to recognize the wrong assumptions hiding in it?

I have been on the experienced side of that comparison. My concern is not that younger people learn quickly. Good for them. My concern is that an experienced worker who refuses to learn the tool can become less visible at the very moment the tool makes other people's work more visible.

AI can summarize a manual, organize messy notes, draft a proposal, and compare possible answers. It can help a new employee get oriented. It cannot take responsibility for a decision, recognize every bad source, earn a customer's trust, or remember why an apparently sensible idea failed three years ago. The most useful combination is experience plus a willingness to learn. That combination is available at 40, 60, or 70.

What the age and AI numbers really show

Pew Research Center tracked whether Americans had ever used ChatGPT. In 2023, 33% of adults ages 18–29 reported having used it, compared with 13% of those 50–64 and 4% of those 65 and older. By 2026, the figures were 61%, 37%, and 19%, respectively. Use rose in every age group. The gap remained. Adults ages 30–49 went from 21% in 2023 to 55% in 2026. [1]

ChatGPT use rose at every age, but gaps remain.% of U.S. adults who had ever used ChatGPT
  • Ages 18–29
  • Ages 30–49
  • Ages 50–64
  • Ages 65+
ChatGPT use rose at every age, but gaps remainThe question wording changed in 2026. Values are shown in the table immediately below.0%20%40%60%2023202420252026
YearAges 18–29Ages 30–49Ages 50–64Ages 65+
202333%21%13%4%
202443%27%17%6%
202558%41%25%10%
2026*61%55%37%19%

The question wording changed in 2026. Share of U.S. adults who had ever used ChatGPT. Source: [1].

There is an important measurement wrinkle. Pew changed its question wording in 2026 to ask about ChatGPT alongside other chatbots. Read the final year as a useful snapshot, with that change in mind, rather than pretending the four rows are identical measurements. And this is a 2023–2026 series because ChatGPT was released publicly in late 2022. A supposed five-year ChatGPT chart beginning in 2021 would invent history. [1]

What about the workplace? In Pew's separate 2026 question, 41% of employed adults ages 18–29 and 43% of those ages 30–49 reported using chatbots for work tasks. Among employed adults 50–64 it was 29%, and among those 65 and older it was 18%. People in their 30s and 40s are hardly outside this story. That matters if you are 40 and assume you can wait until retirement approaches. [2]

Who reported using chatbots for work tasks in 2026?Share of employed adults in each age group; not intensity or productivity
Reported chatbot use for work tasks by age group in 2026Percentages are among employed adults in each age group. The table immediately below repeats all four values.0%10%20%30%40%50%Ages 18–2941%Ages 30–4943%Ages 50–6429%Ages 65+18%
Age groupUsed chatbots for work tasks
Ages 18–2941%
Ages 30–4943%
Ages 50–6429%
Ages 65+18%

Share of employed adults in each age group, 2026. Source: [2].

One study shows why learning the tool can matter. Researchers examined the introduction of a generative AI assistant to 5,179 customer support agents. They estimated that issues resolved per hour rose 14% on average, including a 34% gain for novice and lower-skilled agents, with little impact on the most experienced and highly skilled. The newer workers closed part of a task-specific performance gap. That does not mean a chatbot can replace every senior employee, or that the same percentages will hold in consulting, accounting, medicine, or a machine shop. It does mean that experience alone may become less visible if you never show what you can do with the newer tools. [3]

Age discrimination is already a real concern. AARP Research's 2025 survey found that 64% of workers 50 and older reported seeing or experiencing age discrimination, and 22% felt pushed out because of age. Those are respondents' reported experiences. They do not show that AI caused the bias. Federal age-discrimination law generally protects workers 40 and older, including in hiring, promotion, layoffs, and training. [4]

Here is my prediction, and I want to label it honestly: employers who reward fast, impressive-looking output without checking the thinking underneath could begin undervaluing experienced people earlier in their careers. We do not yet have evidence that the age at which discrimination begins has moved downward because of AI. That is a question to investigate, not a number to put in a headline. In the meantime, you have practical steps you can take.

Your first move: learn AI on work you understand

Do not begin with a $1,000 course or a promise that AI will make you rich. Begin with a real problem from a field you know. ChatGPT has a free option. At the time of writing, ChatGPT Plus is $20 a month, so six months would cost $120 or a year $240 before tax or changes in price. You can start on the free option and pay only when you have a reason to. [6]

Take a public article or a work document your employer permits you to use. Ask: “Explain this in plain English. What decision does it affect? What assumptions are you making? What should I verify at the original source?” Compare its answer with the source. Then ask it to make the explanation shorter. This is practice in directing and checking an assistant, which is much more useful than collecting clever prompts you never use.

Ask it to teach you too. “I know this industry, but I do not know this software. Give me a ten-minute lesson, ask me two questions, and tell me where my answers are weak.” If a manual is hard to follow and you have the right to share it, work through one section at a time. Have the tool explain terms, produce a checklist, and quiz you. Review the actual manual when a step matters.

You can use voice during a commute if you can operate it hands-free and remain fully focused on driving. Ask for a spoken lesson or a set of questions. Do not look at the screen or handle documents while driving. Voice access and limits vary by plan and account. [7]

I would set up two ChatGPT Projects: one for work learning and one for personal learning. In the work project's instructions, write: “Be a demanding business analyst. Challenge my assumptions. Separate facts from guesses. Cite sources when you research, and ask what evidence would change the recommendation.” In the personal project: “Teach patiently in plain English. Give me one small action to try, then quiz me.” These are separate spaces within one account, not two paid subscriptions or a special voice setting. Projects can keep relevant chats, files, and instructions together, subject to plan and workplace settings. [8]

Be careful with private material. Follow your employer's AI policy. Do not paste customer records, medical information, confidential financials, or proprietary manuals into a personal account without permission. Check your account's data controls. Above all, check the finished work. An AI answer can be neat, quick, and wrong. [9]

Make your experience visible again

Pick one task you already perform well. Document how long it normally takes, where mistakes tend to occur, and what a good result looks like. Use AI on the parts that suit it: organizing notes, generating a first outline, comparing alternatives, or finding missing questions. Then apply your judgment. What did the tool miss? What recommendation would have caused a problem? What did your customer or team actually need?

Keep a small before-and-after record. Maybe you cut a two-hour research task to 45 minutes and found two missing assumptions. Maybe you produced a clearer client explanation and reduced follow-up questions. Show the reasoning, not just the polished PDF. Managers should measure accuracy, judgment, and outcomes as well as speed. If they will not, you still have proof of what you can do when you seek a new role or a customer of your own.

Younger coworkers can teach you shortcuts. You can teach them context, judgment, and the questions a model will not think to ask. I would rather have that exchange than a contest over who looks smartest in a meeting.

The second move: build options beyond one paycheck

The conventional model says you work one job, save what you can, and draw on savings if the job disappears. Emergency savings still matter. But when a job search takes longer than expected, every month without income eats into the cushion. The Federal Reserve found that 55% of U.S. adults had put aside three months of emergency expenses in 2025, compared with 59% in 2021. Another 30% of adults said they could not cover three months of expenses through savings, borrowing, or selling assets. These are figures for adults generally; they do not establish that AI caused anyone's financial position. [5]

Adults with three months of emergency savingsSelf-reported national adult measure; not older workers or AI impact
Adults with three months of emergency savingsSelf-reported percentages for U.S. adults generally. The table immediately below repeats both values.0%20%40%60%59%202155%2025
YearAdults with three months of emergency savings
202159%
202555%

Share of U.S. adults, self-reported. Source: [5].

My goal is to build toward at least four sources of income over time, perhaps more where they make sense. Four is a planning rule of mine, not a scientifically proven requirement. I do not mean four jobs. A salary, one defined project service, a recurring service for the same kind of customer, and a modest digital product or paid workshop could be four sources. Two strong, profitable sources are better than seven ideas that never collect a dollar.

There is another trap: four products sold to one customer or through one social account may all fail at once. Watch your dependence on the same buyer, platform, industry, or algorithm. Track cash collected, delivery time, direct costs, taxes, and customer concentration. Gross sales are not take-home income. If three $500 projects produce $1,500, but cost $300 directly and take 30 hours, that leaves $1,200 before tax, overhead, and paying yourself for those hours. Those numbers are only an example; they show why you must do the math.

Where could an additional offer come from? Look at the questions people already bring you. Maybe you repair equipment, run payroll, train a team, care for a property, teach a craft, or know how to evaluate a complicated purchase. Start with the problem people repeatedly ask you to solve. Talk to potential buyers before writing an ebook, buying ads, or building a site. AI makes it easier to draft and organize an offer. A customer deciding to pay is the test.

Free places to learn and get help

You do not have to learn this alone, and free does not mean worthless. Check eligibility and current class schedules before you sign up.

Choose one learning resource and one mentor or practical helper. You do not need a folder of 30 course certificates before you try a real task.

Your 30/60/90-day plan

This is a working schedule, not a promise of income by a deadline. Keep your current paid work if you can. Set aside a few blocks a week for learning and a few for customer conversations. If a stage takes longer, use what you learned and continue.

Days 1–30

Prove a small offer

Learn with an approved work task, talk with potential buyers, and test a narrowly scoped paid pilot.

Checkpoint: a validated buyer conversation, a measured AI-assisted task, and ideally one paid test.

Days 31–60

Make useful work repeatable

Improve the offer from pilot feedback, document delivery, and make another serious sales attempt.

Checkpoint: a repeatable process and evidence for or against recurring demand.

Days 61–90

Build a trustworthy home

Test one related workshop, guide, template, or service extension and review its economics.

Checkpoint: one proven offer, documented delivery, and one related extension tested with buyers.

These are planning milestones from the article, not research findings or a promise of income.

Days 1–30: learn the tool and prove a small offer

Days 1–7: Make a list of three tasks you know well and three problems people ask you to solve. Use ChatGPT on one approved work task each day. Record time spent, what it got right, what you corrected, and what became easier. Set up your work and personal Projects. Take a free Senior Planet or OpenAI Academy beginner lesson.

Days 8–14: Pick one type of customer and one narrow problem. Interview about ten potential buyers. Ask what is going wrong, how they handle it now, what it costs them, and whether they have paid to solve it. Write down their words. Ask AI to group repeated complaints; check the summary against your notes. Do not pitch them an unbuilt platform.

Days 15–21: Put one offer into a sentence: “For [price], I will deliver [specific result] for [buyer] within [time].” Calculate the likely hours and costs. Make one sample or a short proposal. Check any employer outside-work rules and the licenses, registrations, or insurance appropriate to the activity. Ask suitable prospects for a small paid pilot. One to three paid pilots is an illustrative target, not a forecast. [15]

Days 22–30: Deliver the first pilot carefully. Track time, expenses, customer questions, and actual outcome. Ask for honest feedback and permission before sharing any testimonial. If nobody buys, revisit the problem, buyer, scope, or price. Your day-30 checkpoint is a validated buyer conversation, a measured AI-assisted task, and ideally one paid test. Do not buy a package of subscriptions to compensate for missing demand.

Days 31–60: make the useful work repeatable

Days 31–45: Improve the offer using the pilot feedback. Write a simple delivery checklist: intake, work, human review, handoff, invoice, follow-up. Let AI draft routine material, while you keep the final review wherever an error could hurt a customer. Track leads, proposals, invoices, and expenses in one place. Try the revised offer with another customer. A second sale matters more than a glamorous launch page.

Days 46–60: Ask customers what help they need after the original project. If they truly need regular support, test a clearly scoped monthly service. Spell out exactly what is included and how much time it takes. Track whether a customer renews. By day 60, your checkpoint is a repeatable process, another serious sales attempt, and evidence for or against recurring demand. If the first service loses money, fix it before adding more products.

Days 61–90: build a trustworthy home for the offer

Days 61–75: Turn one repeated customer question into a short workshop, guide, template, or second service. Show a sample and ask for a purchase before producing a large course. Keep rights to your own materials; do not reuse an employer's documents. Decide whether your working offer needs a simple website now. The site should answer buyer questions, explain the scope and location or service area, show genuine proof with permission, and make contacting you easy.

Days 76–90: Compare collected revenue, expenses, hours, repeat business, and customer feedback. Keep the offers with evidence of demand and a reasonable return for your time. Test the website's phone and inquiry form yourself. Ask how each inquiry found you. By day 90, aim for one proven offer, a documented way to deliver it, and one related extension tested with buyers. Four lasting income streams may take much longer, and that is fine.

Why a website is now a reputation check

A website should do more than display a logo, phone number, and a generic paragraph. A cautious buyer wants to know what you actually do, who you serve, what the process costs or how a quote works, and why you can be trusted. Put that information in clear pages with current contact details, useful examples, and a straightforward next step. Keep it accurate as your offer changes.

Search and AI tools are part of discovery, but they do not hand out guaranteed leads. Google says its AI search features still depend on sound search basics: indexable pages, accessible text, useful information, and good site quality. OpenAI says a site can be eligible for ChatGPT search when its search crawler can access it; eligible businesses can also claim and update a business profile. Claiming a profile or building a site does not guarantee a recommendation, placement, or sale. What you can control is the quality and clarity of your evidence, and whether a customer can reach you. [16]

You can test the first offer yourself. When customers are paying and you need a site that explains the offer and captures inquiries, hiring a professional can save time and mistakes.

The real starting point is smaller than a website. Open an AI tool. Give it a real task. Check what it tells you. Then talk with one person who might pay for something you know how to do. Experience still counts. Now is the time to make it count in more than one place.

Three books if you want to read further

These are three useful books for different parts of the job, not a statistically ranked “top three”: Ethan Mollick's Co-Intelligence: Living and Working with AI for working with AI while retaining judgment; Kerry E. Hannon's Never Too Old to Get Rich for midlife entrepreneurship; and Paul Jarvis's Company of One for building a sustainable small business. The free resources above are enough to begin. [17]

Sources and reading notes

[1] Pew Research Center, “Americans and AI 2026: Appendix: Detailed chart and tables,” June 17, 2026. ChatGPT ever-use age data for 2023–2026; question wording changed in 2026. https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-appendix-detailed-chart-and-tables/

[2] Pew Research Center, “How Americans' opinions and use of AI differ by age,” June 17, 2026. Table on chatbot use for work tasks among employed adults. https://www.pewresearch.org/internet/2026/06/17/how-opinions-and-use-of-ai-differ-by-age/

[3] Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, “Generative AI at Work,” NBER Working Paper 31161, 2023 revision. This working-paper version reports 5,179 agents, 14% average gain and 34% for novice/lower-skilled agents. Do not silently substitute the later journal version's different sample and estimates. https://www.nber.org/papers/w31161

[4] AARP Research, “Age Discrimination Persists Among Older Workers,” updated July 22, 2026, reporting 2024 and 2025 survey responses. https://www.aarp.org/pri/topics/work-finances-retirement/employers-workforce/age-discrimination-workplace/ ; U.S. Equal Employment Opportunity Commission, “Age Discrimination.” https://www.eeoc.gov/age-discrimination

[5] Federal Reserve, “Economic Well-Being of U.S. Households in 2025: Savings and Investments,” May 2026. Three-month emergency savings for 2021 and 2025; other ways to cover expenses. https://www.federalreserve.gov/publications/2026-economic-well-being-of-us-households-in-2025-savings-investments.htm

[6] OpenAI Help Center, “What is ChatGPT Plus?” Price checked November 20, 2026. https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus

[7] OpenAI Help Center, “ChatGPT Voice,” availability and plan limits. https://help.openai.com/en/articles/20001274-chatgpt-voice

[8] OpenAI Help Center, “Projects in ChatGPT,” instructions, files and availability. https://help.openai.com/en/articles/10169521-projects-in-chatgpt

[9] OpenAI Help Center, “Data controls in ChatGPT.” https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt

[10] Senior Planet from AARP, class listings and free technology help. https://seniorplanet.org/classes/ ; https://seniorplanet.org/tech-help

[11] OpenAI Academy, “ChatGPT for Older Adults” and “AI Foundations.” https://academy.openai.com/public/content/collections/chatgpt-for-older-adults?slugs=chatgpt-for-older-adults ; https://academy.openai.com/public/courses/ai-foundations-juzjs

[12] AARP Foundation, “Digital Skills Ready@50+.” https://www.aarp.org/aarp-foundation/find-help/digital-skills.html

[13] U.S. Small Business Administration, “Counseling” and “Resource Partners,” including SCORE business mentors and SBDC counseling. https://www.sba.gov/counseling/ ; https://www.sba.gov/local-assistance/resource-partners/score-business-mentoring

[14] U.S. Department of Labor, “Senior Community Service Employment Program,” eligibility and training. https://www.dol.gov/agencies/eta/seniors/

[15] U.S. Small Business Administration, “Launch your business,” on registration, permits and related setup. https://www.sba.gov/business-guide/launch-your-business

[16] Google Search Central, “AI features and your website.” https://developers.google.com/search/docs/appearance/ai-features ; OpenAI Help Center, “Searching the web with ChatGPT.” https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt ; OpenAI, “Businesses in ChatGPT.” https://openai.com/business/businesses-in-chatgpt/

[17] Publisher pages: Ethan Mollick, Co-Intelligence, Penguin Random House, https://www.penguinrandomhouse.com/books/741805/co-intelligence-by-ethan-mollick/ ; Kerry E. Hannon, Never Too Old to Get Rich, Wiley, https://uat.store.wiley.com/en-us/never-too-old-to-get-rich-the-entrepreneur%27s-guide-to-starting-a-business-mid-life-p-9781119547907 ; Paul Jarvis, Company of One, Penguin Business, https://www.penguin.co.uk/books/312524/company-of-one-by-jarvis-paul/9780241470466