Practical Ways to Use AI in Your Marketing Without Losing Your Voice

  1. Blog
  2. Ops, Teams & AI

By Saroj Jha · February 19, 2026 · 10 min read

Which marketing jobs to hand to AI and how to brief each one, the six checks a person runs before anything goes out, and the FTC and EU AI Act rules that apply now.

From this article

AI can’t protect a voice you haven’t defined.
AI can scale a documented voice and will happily average an undocumented one — so the work that makes AI safe to use is written down before any tool is introduced.Source: AAJ brand voice framework

Most marketing teams already use AI. In a 2025 survey of 1,015 B2B marketers, 95% said their organizations use AI-powered applications[1]. Across the US economy, 19.8% of businesses said they were using AI in May 2026[2], and among firms that use it, sales and marketing is the most common function[3]. Yet fewer than 4 in 10 of those B2B marketers say AI has improved their performance[4].

In my view, the gap is rarely the tool. It’s which jobs get handed over, and whether a person checks the output before it goes out. This is the practical side: which marketing jobs to give AI, how to brief each one, and the checks and disclosure rules that apply now. Why AI drafts drift toward generic copy in the first place is covered in Why AI-written content doesn’t sound like you.

Write your voice down, then hand AI the jobs where a mistake is cheap to catch: research summaries, first drafts, repurposing and test variants. Keep a named person on every fact, promise and customer quote, working from a written checklist. That review is also what the rules expect: the FTC bans AI-generated fake reviews, and from 2 August 2026 the EU requires a label on AI-written public-interest text unless a person has reviewed it and takes editorial responsibility.

In three numbers

  • 95% of 1,015 B2B marketers surveyed in 2025 say their organizations use AI-powered applications[1]
  • 52% of US firms using AI use it in sales and marketing, the most common business function[3]
  • 19 points fewer right answers from consultants using AI on a task just outside its abilities[5]

Write the voice down before AI touches it

AI can scale a voice that’s written down, and it will average one that isn’t. Before any tool sees a brief, put four things on one page:

  • Personality: three or four traits, such as direct, warm, expert.
  • Words: the phrases you always use and the ones you never use.
  • Sentence style: short and plain, or longer and detailed.
  • Point of view: what you believe, what you don’t, and who you’re for.

Then train the tool on your own work. Paste three to five of your best pieces, ask it to describe the tone, structure and recurring phrases, and edit that description into rules you agree with. Save the result as a reusable instruction, such as a project file or custom instruction, so nobody retypes it. The Brand Voice Playbook covers the full method.

Running case: Example Co. Fictional

Example Co. sells property-management software to small landlords. It has nine people and one marketer. Its voice file opens: “You are writing for Example Co. We are practical, plain and warm. Short sentences. Say ‘landlords’, not ‘users’. Never write seamless, game-changing or leverage. Use no number that isn’t in the source I give you. Write like one landlord talking to another.”

Figure 1 AI can scale a voice only once it’s written down

Before No voice file

Request Write a LinkedIn post about the new reminder feature. Draft

“In today’s fast-paced world, seamless rent collection is a game-changer. Leverage automated reminders to cut late payments by 30%!”

  • Category filler
  • Two words on the never list
  • A number nobody gave it

After With Example Co.’s voice file

Request The same request, plus the voice file and the feature notes. Draft

“A late rent payment costs you a week of chasing. The new reminders text tenants three days before rent is due, and you see who has paid.”

  • No banned words
  • Speaks to landlords, never ‘users’
  • Only facts from the feature notes

Illustrative: both drafts were written for this figure, and Example Co. is fictional. The point is the article’s own: AI can’t protect a voice you haven’t defined.

Do this

Write the one-page voice file, add three before-and-after rewrites from your own copy, and load it into every AI tool your team uses before anyone drafts with it.

Which marketing jobs to hand to AI

Hand AI the jobs where a wrong answer is cheap to spot and fix. Keep people on the jobs where a wrong answer reaches a customer as a fact or a promise. The research supports that split.

In a preregistered experiment, 453 college-educated professionals, marketers among them, did writing tasks from their own occupations. Those given ChatGPT took 40% less time, and their work was rated 18% better[6]. The tasks needed no company knowledge and no fact-checking, which is most of what real marketing work needs. A field experiment with 758 Boston Consulting Group consultants found the other edge. On tasks within GPT-4’s abilities, those using it finished 12.2% more tasks, 25.1% faster, at more than 40% higher quality. On a task just outside them, they were 19 percentage points less likely to reach the right answer[5].

Figure 2 Inside AI’s abilities, writing got faster and better; just outside them, answers got less accurate

Change for people using AI, against people without it (percent)

  • Less time per writing taskNoy and Zhang, 453 professionals 40% Speed
  • Higher quality ratingNoy and Zhang, 453 professionals 18% Quality
  • More tasks completedBCG consultants, tasks inside GPT-4’s abilities 12.2% Volume
  • Faster to completeBCG consultants, tasks inside GPT-4’s abilities 25.1% Speed
  • Higher qualityBCG consultants, inside GPT-4’s abilities over 40% Quality

On a task just outside GPT-4’s abilities (percentage points)

Fewer right answersBCG consultants, outside −19 pts Accuracy

Neither study tested brand voice. Noy and Zhang’s tasks needed no company knowledge and no fact-checking. That is the part of marketing a person still has to do.

Show the numbers as a table
Study Measure With AI, against without
Noy and Zhang (2023) Time per writing task 40% less
Noy and Zhang (2023) Quality rating 18% higher
Dell’Acqua et al. (2023), inside Tasks completed 12.2% more
Dell’Acqua et al. (2023), inside Speed 25.1% faster
Dell’Acqua et al. (2023), inside Quality More than 40% higher
Dell’Acqua et al. (2023), outside Right answers 19 percentage points fewer

Noy and Zhang (2023), Science; Dell’Acqua et al. (2023), Harvard Business School Working Paper 24-013, 758 Boston Consulting Group consultants. Both sets of bars are drawn on the same 0 to 40 scale; the last bar is in percentage points.

So AI is strongest at drafting and reshaping, and weakest where it sounds most sure of itself: facts, claims and judgment. Figure 3 sorts common marketing jobs along that line.

Figure 3 Hand AI the jobs where a mistake is cheap to catch; keep people on facts, promises and customers
  • Marketing job What AI does What a person checks Verdict
  • Editing Shortens, clarifies, offers headlines The meaning hasn’t changed and nothing new was claimed Hand over
  • Repurposing Turns one source piece into posts, emails and scripts Everything in the new piece is already in the source Hand over
  • Test variants Writes many subject lines, headlines and opening lines Each passes the voice rules and claims nothing beyond the base copy Hand over
  • Research summaries Summarizes reports, groups call notes into themes, maps competitor messaging Every number and quote, against the original With checks
  • First drafts Angles, outlines and drafts from your brief and voice file Facts, examples and opinions, added or verified by a person With checks
  • Positioning, pricing and offers Options and notes only A person writes and decides Keep human
  • Crisis and sensitive messages Talking points only A person writes and signs Keep human
  • Customer quotes, reviews, testimonials Nothing: never generated Real, verified and used with permission Keep human
  • Confidential customer data Only in tools whose data terms you’ve checked A person decides what goes in Keep human
  • Hand over AI does it; a person spot-checks
  • With checks a person verifies every fact
  • Keep human a person writes and owns it

My grouping, based on the evidence in Figure 2 and the rules in this article; not a tested ranking.

Do this

List the marketing jobs your team repeats every month, mark each “hand over”, “hand over with checks” or “keep human”, and start with one or two from the first group.

Research: a fast summary, never the source

AI is good at reading a pile of material quickly: summarizing long reports, grouping customer complaints into themes, mapping how competitors describe themselves. It is not a source. A model will produce a plausible statistic or citation that doesn’t exist, which is why Google’s guidance for site owners says “It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing”[7].

Give it your own material to work from, such as call notes, reviews and survey answers, and make it show its evidence. For example: “From these 20 call notes, list the five most common frustrations. For each, quote two lines word for word and say how many notes mention it.”

Do this

For any number AI gives you, open the original source, check the figure, the date and the sample, and cite that source. If you can’t find it, don’t use the number.

First drafts: you choose the angle, AI does the groundwork

Use AI for angles, outlines and first drafts, not final copy. Ask for options before you ask for prose: “Give me 10 angles on [topic] for [audience], in a practical, plain tone.” Pick two or three, combine them, and only then ask for an outline built on your angle and your own examples.

The person adds what the model can’t know: your customer stories, your numbers, your opinions. Then use AI again as an editor, against your rules: “Make this paragraph shorter and clearer; keep the meaning and the tone,” or “Offer 10 headlines for this article. No clickbait; lead with the benefit.”

Do this

Brief AI the way you’d brief a freelance writer, with the audience, the angle, the source material, the voice file and the length. A one-line prompt gets a generic draft.

Repurposing: one human piece, many formats

Start from one piece a person made, such as a webinar, an article or a customer interview, and let AI turn it into social posts, emails, ad copy and short video scripts. Every derivative comes from the same source, so the voice and the facts stay consistent, and the check is simple: everything in a derivative must already be in the source.

Running case: Example Co. Fictional

Each month Example Co. runs a 40-minute webinar for landlords. Its marketer used to spend about 10 hours turning it into a recap article, six LinkedIn posts and two emails. Now AI drafts all nine pieces from the transcript and the voice file, and she spends about 4 hours, 1.5 of them on checks. In the first month the check caught a LinkedIn draft saying the product “cuts late rent by 30%”. Nobody had said that in the webinar, so she cut it. All figures are illustrative.

Figure 4 Example Co.’s marketer cut repurposing from 10 hours to 4, and kept 1.5 of them for checks

Before Writes all nine pieces herself

10 h

With AI AI drafts from the transcript; she edits, then checks

2.5 h 1.5 h
  • Writing by hand
  • Editing AI drafts
  • Checks against the source

What the checks caught in month one: a LinkedIn draft saying the product “cuts late rent by 30%”. Nobody said it in the webinar, so it was cut.

Show the numbers as a table
Monthly repurposing Writing by hand Editing AI drafts Checks Total
Before 10 h 0 h 0 h 10 h
With AI 0 h 2.5 h 1.5 h 4 h

Fictional: Example Co.’s monthly webinar turned into one recap article, six LinkedIn posts and two emails. Bars are drawn to the same 10-hour scale.

Do this

Keep one repurposing prompt that names each channel’s format and length, and add the rule “Use only facts, numbers and quotes that appear in the source. If something is missing, write [CHECK].”

Testing variants: AI writes the options, your data picks

AI is good at producing many versions of short copy: subject lines, ad headlines, opening lines for different roles. Use it to widen the options, then let a real test choose. Give the model the base copy and the rules: “Using this email, write five opening lines for landlords with one or two units and five for landlords with 10 or more. One or two sentences each, no flattery, and no claim that isn’t in the email.”

Personalize from what you actually know, such as role, industry or the page someone downloaded, never from details the model guesses. Don’t paste confidential customer data into a tool whose data terms you haven’t checked. Test one change at a time and wait for enough volume before you call a winner; AAJ’s A/B Test Significance Calculator tells you when a difference is likely to be real.

Do this

Have AI write 10 variants, cut them to the two or three that pass your voice rules, and test those against your current version.

The person-checks step

Every AI-assisted piece gets a named person who reads it before it goes out, against a written checklist. Without the checklist, review shrinks to a skim. Six questions cover most marketing.

Figure 5 Six checks a named person runs before any AI-assisted piece goes out
  1. 1 Is every number, date and quote traced to a source you opened? Example Co.: “cuts late rent by 30%” was in no source Yes If no Find the source or cut it
  2. 2 Can you prove every claim and promise? “Set up in a day” holds only for landlords with under 10 units Yes If no Narrow the claim or remove it
  3. 3 Is every customer quote, review and testimonial real and used with permission? No quote was written by a tool Yes If no Remove it; the FTC rule bans fake ones
  4. 4 Does it pass your voice rules? No banned words; “landlords”, not “users” Yes If no Edit it to the voice file
  5. 5 Did only approved data go into the tool? Call notes went into a team plan with training switched off Yes If no Stop and tell the owner of your AI policy
  6. 6 Are the required labels in place? The chatbot says it’s AI; no deepfakes in the ad Yes If no Add the label before it goes out

All six are yes: publish, and log the reviewer’s name

My checklist. The rules behind questions 3 and 6 are in “The disclosure rules that apply now”; Example Co.’s answers are fictional.

Some work stays human from the start: positioning and brand story, pricing and offer promises, crisis and other sensitive messages, and anything involving confidential customer data. AI can draft notes and options for these, but a person writes and owns the message. AAJ’s AI Tool Policy for Marketing puts these rules on one page your team can agree to.

Do this

Name a reviewer for each type of content, paste the six questions into your approval step, and log what the checks catch for a month. The log tells you which jobs AI can take more of.

Free tool

Check the claims before a person signs off

The free Claims & Proof Checker flags the claims and statistics in a draft that need a named source, so the reviewer starts with a list instead of a blank page.

Open the checker

The disclosure rules that apply now

Two sets of rules matter most for marketing teams selling in the US and the EU, and both reward the person-checks step. This is general information, not legal advice.

FTC: no fake reviews, AI-written or not

The FTC’s rule on consumer reviews and testimonials, announced in August 2024 and in force since 21 October 2024, bans reviews and testimonials that misrepresent that they are by “someone who does not exist, such as AI-generated fake reviews”. It also bans undisclosed reviews by company insiders and rewards for reviews with a set sentiment, and it lets the agency seek civil penalties from knowing violators[8]. In December 2025 the FTC sent warning letters to 10 companies about possible violations[9]. So never let a tool write a customer quote, review or testimonial.

EU AI Act: labels from 2 August 2026

Article 50 of the AI Act applies from 2 August 2026[10]. Three parts matter for marketers. A chatbot you build or put your name on must tell people they are dealing with AI, unless that’s obvious. Deepfakes, meaning AI-generated or manipulated images, audio or video that resemble real people, places or events and would falsely appear authentic, must be disclosed. And AI-generated text published to inform the public on matters of public interest must be labelled, unless a person has reviewed it or exercised editorial control and someone holds editorial responsibility. The Commission says spell-checks and other superficial checks don’t count as review.

The Digital Omnibus on AI, Regulation (EU) 2026/1744, gave providers of generative AI systems already on the market until 2 December 2026 to mark their outputs in a machine-readable way. It did not delay the duties on businesses that use those tools[10]. The Commission’s voluntary Code of Practice on marking and labelling AI-generated content, published in June 2026, sets out how to apply the labels[11].

Search and copyright

Google’s guidance warns that using generative AI “to generate many pages without adding value for users” may violate its spam policy on scaled content abuse[7]. The US Copyright Office concluded in January 2025 that AI outputs can be protected “only where a human author has determined sufficient expressive elements,” not by “the mere provision of prompts”[12]. A person’s real edits are what make the work yours.

Figure 6 The rules that touch AI-assisted marketing, in the order they arrived
  1. Aug 2024 FTC final rule on fake reviews and testimonials Bans reviews by “someone who does not exist, such as AI-generated fake reviews”; in force from 21 October 2024.
  2. Jan 2025 US Copyright Office report, part 2 Prompts alone don’t make AI output protectable; a human author’s expressive choices can.
  3. Dec 2025 FTC warning letters 10 companies warned about possible violations of the review rule.
  4. Jun 2026 EU Code of Practice on marking and labelling Voluntary; sets out how to label deepfakes and AI-written public-interest text.
  5. 2 Aug 2026 EU AI Act Article 50 applies Chatbots say they’re AI; deepfakes are disclosed; unreviewed AI text on matters of public interest is labelled.
  6. 2 Dec 2026 Marking grace period ends Generative AI systems on the market before 2 August must mark outputs in a machine-readable way.

FTC (2024, 2025); US Copyright Office (2025); European Commission (2026), Article 50 FAQ and press release IP/26/1328. General information, not legal advice.

Do this

This quarter, remove any AI-written review, quote or testimonial from your site, ads and sales decks; add a line saying “you’re chatting with an AI assistant” to any chatbot you run; and name the person who reviews and signs off each AI-assisted article you publish to EU readers.

Start small and measure what changes

Pick one or two jobs from the first group in Figure 3 and run them for a month. Track four things: hours per piece, including review time; what the checks caught; engagement quality, such as replies, saves and comments rather than impressions; and lead quality rather than lead count. More output isn’t the goal: in the same B2B survey, 12% of marketers said the quality of their content decreased with AI[1]. If AI-assisted pieces do worse or read off-brand, fix the voice file, the brief or the checks before you add more.

Write down what works as you go: the prompts, the voice file, the checklist and who approves what. That becomes your team’s AI playbook, and it’s what lets a new hire or a freelancer use AI without drifting from your voice.

Do this

Choose the jobs, set the four measures, and review them after four weeks before you hand AI a third job.

How do you choose AI tools for SaaS content marketing?

AAJ doesn’t rank or recommend AI tools, because the right one depends on your content jobs, your data and the stack you already run. Start from the job, not the tool. Match each job in Figure 3 to a kind of tool: a general-purpose AI assistant, a writing tool built on one, an SEO content tool, a design or video tool, or an AI feature in software you already pay for (check that first).

Judge every candidate on the same criteria

  • Output quality on your own content, not the vendor’s demo.
  • Voice control: can you load your style rules and banned words?
  • Data terms: do your inputs train models, how long are they kept, and what admin controls exist?
  • Output rights: what the terms say about ownership.
  • Integrations with your CMS, docs and approval workflow.
  • Cost model and how it scales with usage.
  • Human review built into the workflow.

Run a short trial

Give two or three candidates the same real briefs, score each against the criteria, and count the editor’s time to fix each output. Set a decision date first.

By stage

At Seed, I would start with one general-purpose assistant plus the AI features already in your stack, and invest in a written voice file. At Series A–B, more writers and more volume make data terms, admin controls, integrations and cost scaling matter more, so run the formal trial.

The common mistake is buying a tool before defining the job and the voice. AI then makes you faster at generic content.

If you want me to look at your own funnel, positioning and channels, my Growth Audit ends with three prioritized fixes and a 30-day action plan.

Related on AAJ

Articles

Playbooks

Templates and tools

References

The superscript numbers in the article match this list.

Studies, regulations and web pages were checked on October 21, 2026.

  1. Rose, R. (2025). 9 Takeaways and Insights From the 2026 B2B Content and Marketing Trends Report. Content Marketing Institute and MarketingProfs, October 8, 2025. Survey of 1,015 B2B marketers, mostly in North America, fielded June 24 to August 14, 2025. contentmarketinginstitute.com
  2. Grundy, A., Breaux, C. and Khatiwoda, D. (2026). Large Firms With at Least 20 Employees Biggest AI Users. US Census Bureau, May 26, 2026. census.gov
  3. Bonney, K., Breaux, C. L., Dinlersoz, E., Foster, L. S., Haltiwanger, J. C. and Pande, A. A. (2026). The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks. NBER Working Paper 35141, using the US Census Bureau’s Business Trends and Outlook Survey AI supplement, November 2025 to January 2026. nber.org
  4. Content Marketing Institute (2025). New B2B Research Finds Winning Marketing Teams Are Building Fundamentals, Not Just Playing With Prompts. Press release, October 8, 2025. contentmarketinginstitute.com
  5. Dell’Acqua, F., McFowland III, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F. and Lakhani, K. R. (2023). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Working Paper 24-013, September 22, 2023. 758 Boston Consulting Group consultants. mitsloan.mit.edu
  6. Noy, S. and Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science 381(6654), 187–192. 453 college-educated professionals, half given ChatGPT. doi.org
  7. Google Search Central (2026). Google Search’s guidance on using generative AI content on your website. Last updated October 1, 2026. developers.google.com
  8. Federal Trade Commission (2024). Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials. Press release, August 14, 2024; the rule (16 CFR Part 465) took effect October 21, 2024. ftc.gov
  9. Federal Trade Commission (2025). FTC Warns 10 Companies About Possible Violations of the Agency’s New Consumer Review Rule. Press release, December 22, 2025. ftc.gov
  10. European Commission (2026). Transparency obligations under Article 50 AI Act. Frequently asked questions, last updated July 24, 2026. digital-strategy.ec.europa.eu
  11. European Commission (2026). Press release IP/26/1328 on the final Code of Practice on marking and labelling AI-generated content, June 10, 2026. europa.eu
  12. US Copyright Office (2025). Copyright Office Releases Part 2 of Artificial Intelligence Report. NewsNet Issue 1060, January 29, 2025. copyright.gov

More in Brand & Voice

Part of the Brand & Voice hub - see all 11 resources on this topic.

Also useful in Marketing Ops, AI & Team.