AI skills on a resume: tool, task, and an artifact you can show

Most AI skills lines fail for a boring reason: they describe a feeling about a product. “Proficient in ChatGPT,” “skilled in generative AI,” and “AI-powered professional” tell Workday nothing it can file and tell a hiring manager nothing they can test. Anyone with a free account can paste that sentence. The screen starts when someone says “show me the last thing you shipped with it.” If you cannot open a doc, a ticket, a deck, a sheet, or a pull request, the line is decoration.

Write three parts. The tool is the product name as the employer spells it: ChatGPT, Microsoft Copilot, GitHub Copilot, Gemini, Claude, Midjourney. The task is the work you actually did: first draft of a help article, weekly SKU exception table, subject-line variants, unit-test stubs. The artifact or volume is what you can put on a screen: 14 Zendesk articles you edited and published, a Monday digest for six stores, eight Klaviyo variants with a 1,200-name holdout, twelve merged tests. That triad is the skill. The adjective is not.

Prompt engineering is a method, not a job title you award yourself after a weekend course. If you wrote structured prompts, say what they produced: a 9-page SOP outline you then rewrote with ops, a macro pack a QA lead approved, a SQL draft a analyst corrected. “Prompt engineer” on a customer-support resume reads like a costume. Put the method inside the bullet if the posting used that noun and you can walk through one prompt, one edit, and one shipped file. For the difference between a duty and a result you can defend, stay with resume achievements. This page stays on how the AI noun earns a place.

What counts as an artifact you can show

• A named file you still have: help article, SOP, briefing, email send, slide deck, sheet, dashboard tab, pull request.
• A volume you held without a secret dashboard: articles per month, tickets summarized per shift, variants per campaign, stores on a digest.
• A human gate you can name: editor, QA lead, pharmacist, store manager, staff engineer who reviewed the output before it shipped.
• Not a chat log. Not “I use it every day.” Not a certificate PDF with no shipped work attached.

Volume is the honest fallback when Legal will not let you show the file. “Summarized 15–20 Zendesk tickets per shift into a handoff note I posted in Slack” is still tool + task + volume. “Used AI to improve communication” is not. If you cannot name the queue, the tool, and the note, cut the line. The same honesty rule as the rest of resume skills applies: if you cannot explain it on a screen, it does not belong on this version of the file.

Put the product name in a Skills heading the ATS already maps: Skills, Technical Skills, or Tools. Group it with the other systems you actually used, not in a special “AI” cloud with icons. “AI tools: ChatGPT. Support: Zendesk, Salesforce Service Cloud.” is a block a recruiter can scan in two seconds and a parser can tokenize. A 90-word paragraph that starts with “I am passionate about the future of generative AI…” is how the noun gets lost between adjectives. The grouping method is the same as any other tool list — see resume skills — you are only adding products you can defend.

Repeat the same noun inside an experience bullet. Greenhouse keyword search often runs on the whole parsed document, but Workday recruiters still filter on structured skills when those fields populated. One mention in a buried summary is weaker than a labeled line plus a bullet that shows the task and the artifact. “ChatGPT” in Skills and “Drafted 14 Zendesk help articles in ChatGPT, then edited and published each one” under Support Specialist is the pair that survives both search and a screen. For which posting nouns to keep, use resume keywords. Do not paste the entire “AI fluency” paragraph from the ad into a footer.

Do not hide the only ChatGPT mention in a two-column sidebar, a logo row, or a header text box. Workday, Greenhouse, Taleo, Lever, and iCIMS extract text and hang it on fields. If the product name lives in a pill graphic, the extract can come back empty. A recruiter then searches “ChatGPT” or “Copilot” and you are not in the set, even though the PDF looks current on your laptop. Use one column. Body 10.5–12 pt. Selectable text. The Enhance CV builder already maps a Skills heading; type the nouns as plain lines.

File type, fonts, and what not to draw

• Upload .docx or a tagged, selectable-text PDF. A Canva “AI skills” badge row exported as an image is a brick.
• Calibri, Arial, Georgia, or the system font in Enhance CV. No script, no ultra-thin display face for the tool line.
• Product names as text. No ChatGPT sparkle icon, no Copilot wing, no skill bar that says “AI 9/10.”
• Dates do not belong next to ChatGPT. Tenure belongs next to the job that used the tool.

Test the file the way the ATS will. Select all in Preview or Word. If you cannot highlight “ChatGPT,” neither can Greenhouse. Then run the ATS checker against the posting so you are not guessing which nouns survived. If you used ChatGPT to draft the resume itself, that is a different workflow — ChatGPT resume that still parses — and it does not replace a proof bullet about work you shipped with the tool.

Hiring managers outside engineering are tired of “AI-powered” as a personality trait. They are not tired of a coordinator who can open the digest, the macro pack, or the email send. Ops, customer support, and marketing already have artifacts. The tool is just the first draft or the first pass. You still own the edit, the publish, and the number. If you only used ChatGPT to rewrite your own resume, that belongs in how you built the file, not under Skills. Work you did for the company belongs on the page.

Operations. Typical honest lines: weekly exception tables, SOP first drafts, vendor-email drafts, shift-note summaries, inventory comments cleaned for a Monday standup. Example shape: “ChatGPT: first pass of a weekly SKU exception digest (store, SKU, qty) for 6 stores; I pasted the output into Google Sheets and corrected every row before standup.” The artifact is the sheet. The volume is six stores. The human gate is you and the standup. “Streamlined operations with AI” hides all three.

Customer support. Typical honest lines: help-center drafts, macro first passes, ticket summaries for handoff, knowledge-base outlines a QA lead approved. Example shape: “ChatGPT: drafted 14 Zendesk help articles and 12 macros; QA approved 9 articles and 8 macros I then published.” Name Zendesk. Name the count. Name the gate. “AI-powered support professional” is not a Greenhouse search term. For queue volume and CSAT rules that still apply when an AI tool is in the line, the support page is customer service resume — this page only adds the tool-task-artifact pattern on top.

Marketing. Typical honest lines: subject-line variants, brief outlines, image-concept prompts a designer picked from, FAQ drafts for a PDP, social captions you rewrote before scheduling. Example shape: “ChatGPT: 8 subject-line and preview-text variants for a Q3 win-back send; I picked 2 after a 1,200-name holdout in Klaviyo.” Midjourney or Firefly only if you actually generated the concepts and a designer or brand lead used one. “Created engaging AI content at scale” is a slogan. The send and the holdout are the artifact.

When the posting is engineering — and when it is not

• Engineering postings that name GitHub Copilot, Cursor, or Claude in the IDE want a pull request, a test file, or a review comment you can open. That write-up lives on software engineer resume.
• Ops, CS, and marketing postings that say “AI fluency” or “ChatGPT” still want the triad. They do not want a fake Copilot line.
• A hybrid coordinator who used ChatGPT for SOPs and Excel Copilot for a real workbook can list both — each with its own task and file.
• If the ad lists five model names and you used one, list the one. Keyword stuffing five LLMs is how resume keywords go wrong.

You do not become an engineer by listing Copilot. You do not fail a marketing screen by listing only ChatGPT if ChatGPT is what you used. Title honesty still matters. “Prompt engineer” is a costume on a coordinator resume. “Operations coordinator — ChatGPT first drafts of SOPs, edited with the store manager” is a job a human can staff. Keep chronology. Keep the real title. Let the bullet carry the tool.

Recruiters skim Skills in seconds, then look for the same noun beside a queue, a campaign, or a store list. “Proficient in ChatGPT” hides the work. Below are three rewrites we use in edits — support, operations, and marketing. Steal the triad. Do not steal the counts. If the bullet still reads like a duty after you place the noun, sharpen the result with resume achievements.

A fourth pattern we cut constantly: a comma list of ChatGPT, Claude, Gemini, Copilot, Cursor, and Midjourney after a two-hour tour of free tiers. Greenhouse may index every token. The interviewer will pick one. If you cannot open a file from that product, delete it. Three to six labeled tools beat a model-name soup. Older experiments go on a master document, not the Workday attachment.

If you have no branded AI tool at work — some floors ban public models — do not invent one. Write the method you did use: templates, macros, a sanctioned internal assistant if you can name it. “Internal assist: first draft of the returns SOP in the company wiki, edited with the DC lead” is still a triad. “Secretly used ChatGPT against policy” is a line you should never put on a resume, even if it is true.

This is an editing sequence, not a theory. Block 45 minutes. Have the job ad, your last resume, and a notes app open to the files you can still find. If you prefer to type inside a known-good layout, start in the Enhance CV builder and paste groups as plain lines. If you want ChatGPT to help you draft, read ChatGPT resume ATS first so the chat does not invent Copilot for you.

1. Save the job ad. Circle every AI product, method, and adjacent system (Zendesk, Klaviyo, Excel, GitHub). Ignore adjectives like “AI-native” and “forward-thinking.”
2. Open a new file (do not overwrite the master). Put contact in the body, then experience, then a heading that literally says Skills.
3. Make a short label: AI tools — or fold the product into Systems if you only have one. Empty “AI” headings look like you forgot a section.
4. Under that label, type only products you used on a real shift, campaign, or project — not a 20-minute vendor demo or a weekend login.
5. Match spelling to the posting when both are true: “Microsoft Copilot,” not “copilot,” and not GitHub Copilot if you only used the Excel add-in.
6. For each kept product, write one messy sentence: tool + task + artifact or volume. Do not polish yet. This is the raw proof list.
7. Turn each sentence into a bullet under the job where the work happened. Name the human gate if you have one (QA, manager, designer, reviewer).
8. Delete any leftover proficiency adjectives and the words “generative AI,” “LLM,” and “AI-powered” unless the posting used them as a required noun you can attach to a file.
9. If the ad lists Copilot and you never used it, write the substitute you did use in the summary in plain language. Do not stuff Copilot into a footer.
10. Mirror each kept noun at least once in Skills and once in a bullet so Greenhouse search and a human see the same word in a story.
11. Export selectable text. Confirm you can highlight every product name. Open the PDF on your phone; if the AI line wraps into a cramped sidebar, flatten to one column.
12. Paste the resume and the job ad into the ATS checker. Fix missing headings and noun gaps, then upload to Workday or Greenhouse.

Step 8 is the one people skip. They keep “AI-powered” because it feels current. It costs a line that could have held “Zendesk” or “Klaviyo.” Recruiters do not search for adjectives. Parsers do not map them to a skills field in a useful way. Step 9 is the ethics gate. A missing Copilot is a conversation. A fake Copilot is a failed screen.

If you stall, you are probably writing the summary first. Write the proof bullets, then a three-line summary that repeats one tool and one artifact. Do not let the chat invent a stack. The ChatGPT resume page is for drafting discipline, not for adding products you have not opened.

Use this as a desk-side filter while you edit. If a choice is not in the table, ask: will Workday store the product name, and can you share a screen in the first ten minutes of a call?

Functional resumes that replace dates with an “AI skills” cloud still show up in career-change templates. Greenhouse can parse some of them; humans distrust them because the calendar disappears. If you are changing fields, keep chronology and let one honest AI bullet sit under the job that used the tool. Do not make the reader reconstruct when you worked. Soft skills stay out of the AI line: put people work in a result bullet, not as “collaborated with AI.”

Course-only ChatGPT. List the course under Education or Projects and name the file you built: “Google IT Support cert — ChatGPT lab: 6-article help-center draft I then rewrote in Docs.” Do not move ChatGPT into Skills as if you used it on a paid queue. A hiring manager will ask which tickets. iCIMS knockout questions and live screens catch the gap faster than the parser will.

The posting names Copilot and you never used it. Say so in the summary with the tool you did use: “ChatGPT for help-center drafts; have not used Microsoft Copilot.” Do not add Copilot to Skills. Do not add it in white text. Do not hope Taleo will not ask. A close substitute you can demonstrate beats a borrowed product name. The same rule applies to Claude, Gemini, Cursor, and Midjourney.

The floor banned public models. Write the sanctioned tool if you can name it, or write the method without a brand: templates, macros, a company assistant. Do not advertise that you used ChatGPT against policy. That is not a skill. It is a risk the employer already decided they do not want.

Students and first jobs. Class projects count when they shipped: “ChatGPT: outline for a 12-slide campaign critique; I rebuilt every slide after the seminar.” Coursework lists that say “exposed to generative AI” are weak. Pair with resume with no experience if the rest of the page is thin. A named artifact still beats a suite you only watched in a lecture.

Engineering screens. If you are applying as an engineer, Copilot or Cursor belongs next to a pull request, a test file, or a review, not next to “used AI to code faster.” Details live with software engineer resume. If you are not applying as an engineer, do not borrow that stack to look technical. Ops and marketing files that list GitHub Copilot without a repo are a common reason screens go cold.

What to do next today: pick one posting, write four labeled skill nouns you can defend, attach a task and an artifact to each AI noun, then score the file. If the checker flags a missing Skills heading or a product name that never appears in selectable text, fix that before you add another adjective. The Enhance CV home page is the shortest path back to the builder and the checker in one place.