What to hand to AI and what people keep
Generative AI is a tool that produces text and images in a short time. In social media work, it is useful for draft post text, ideas for phrasing, illustrations and summaries of long documents. Because it reduces time spent staring at a blank page, it helps most when the person in charge also has other jobs.
But the responsibility for publishing what AI made rests with the company that publishes it. If the text contains a mistake, or the image resembles someone else's work, it is received as the company's message once it goes out from the business account. Before you start, decide what you will hand to AI and what people will keep.
| Task | Easy to hand to AI | People keep |
|---|---|---|
| Post text | Drafts, phrasing ideas, length adjustment | Checking the facts, deciding whether to publish |
| Images | Illustrations, backgrounds, atmosphere | Real photos of products, dishes and completed work |
| Replies | Draft answers to routine questions | Complaints and consultations with individual circumstances |
| Planning | Brainstorming, angles | Deciding what to say |
Copyright: AI output is judged by the same yardstick as any other work
Japan's Agency for Cultural Affairs has set out its view on generative AI and copyright in "General Understanding on AI and Copyright in Japan" (March 2024). It explains that whether using something made with AI infringes copyright is judged, just as if a person had made it without AI, by similarity to and reliance on an existing work.
Similarity means whether creative expression is shared with an existing work. Sharing only an idea or a fact does not count. Reliance means whether the work was made on the basis of an existing work. In the Agency's explanation, a case where the user knew an existing work and had AI produce something similar is given as an example where reliance is likely to be found.
What to watch in social media work
In practice the rule is simple: do not instruct AI to imitate a specific work, artist or character. Asking for "an illustration in the style of so-and-so" or "something like that character" makes a similar result more likely and makes reliance easier to find. For the same reason, avoid pasting another company's ad or another account's post and asking AI to "write it the same way."
Whether things made with AI are protected as your own works is also a point in the Agency's materials. A person needs to have used AI as a tool, with creative intent and a creative contribution. For things you want to protect as yours for a long time, such as a logo or the design of a signature product, it is safer not to leave everything to AI.
AI labels: each network's rules (as of October 2026)
The main networks have labeling rules for content made or altered with AI. What they have in common is that realistic-looking video, audio and people are covered.
YouTube asks creators to disclose AI use at upload for altered or synthetic content that looks realistic. Examples include making a real person say something they did not say, altering footage of a real event, and realistically creating a scene that did not happen. Light uses, such as beauty filters, captions, AI help with scripts and thumbnails, or brainstorming ideas, are listed as examples that do not need disclosure. Repeatedly failing to disclose on purpose can lead to action such as removal.
On Facebook, Instagram and Threads, Meta says it may require an AI label on realistic video and audio created with digital processing that includes AI. Images are outside this requirement, but a label is shown on images that Meta's systems detect as generated or changed by AI. TikTok likewise says scenes or people that look real must be labeled if they were generated or heavily edited with AI.
For business accounts, there are few occasions that call for realistic AI people or scenes in the first place. Use real photos and video where you show the real thing, and keep AI to illustrations and atmosphere, and you will have less to worry about under the labeling rules.
Fact-checking: AI writes plausible mistakes
Generative AI can state things that are not true in convincing sentences. Opening hours, prices, dates, numbers and explanations of laws and systems are especially prone to errors. Always have a person read what AI wrote before publishing, and check the factual parts against the original material.
The quickest way to reduce mistakes is to give AI accurate material. If you hand over your company basics, your products and your wording rules in a document, AI writes from them. If you simply ask "write a post for our company" with no material, AI fills the gap with generalities that fit anyone, or with things that are not true.
Generic text is rarely read
Text that AI writes without material tends to be lines that fit anything, such as "this is important" or "please use this as a reference." Such text does not make readers stop. In X's recommendation system, thin posts are easily met with "Not interested" and are scored much lower. The mechanism is explained in what makes a post show up in X's recommendations. Put at least one fact that only your company could write into every AI draft.
Information you must not enter
How information typed into an AI service is handled differs by service. As a rule, do not enter anything that must not leave the company, such as customer names and contact details, the contents of contracts with business partners, or information on products not yet announced. If you use AI inside the company, decide what may and may not be entered. How to decide is covered in how to write social media guidelines.
What to decide internally
When you build AI into your workflow, a short document covering the following lets anyone keep using it the same way even when the person in charge changes.
- Tasks given to AI and tasks people keep
- Information that must not be entered into AI
- Never instruct AI to imitate a specific work or artist
- Who checks what before AI-made content goes out
- When an AI label is needed and how to add it
Item 4 in particular is least likely to be missed if you set it as an approval flow. How to set one up is explained in how to build an approval workflow.
How SNS TOWN uses AI
SNS TOWN's AI drafts are built on the cautions above. When AI writes, it reads only the company profile, key messages and Library files registered for each company or brand. It is instructed not to fill gaps with generalities and to include that company's facts.
Every draft passes a check before publishing, and only drafts a person has approved are scheduled. Before approving, a person can read the factual parts and fix them. The reference accounts feature reads only the "pattern", such as length and how posts open, from accounts you want to learn from, and is built not to copy their wording.
AI images are drawn as illustrations when the Library has no suitable image, and are instructed not to include text or logos. Real products and completed work are meant to use Library photos. How much you hand to AI can be chosen per post cycle between draft mode (a person approves before publishing) and auto mode, but we recommend starting with draft mode.
Frequently asked questions
Do I have to tell readers that AI wrote a post?
For text, the main networks do not require an AI label across the board. What they ask to be labeled is mainly realistic-looking video, audio and people. Still, however you used it, the company is responsible for what it publishes.
Is it okay to use AI-made images in posts?
There are many situations where illustrations and backgrounds that do not imitate a specific work or character work well. For showing real things, such as products, dishes or completed work, an AI image would show something different from the real thing, so it is safer not to use one.
Can I use another company's posts as a model and have AI write from them?
Avoid pasting another company's text and asking AI to "write it the same way." Using structure such as length and how a post opens as a reference is different from copying wording.
If I leave it to AI, won't every company's posts look alike?
They tend to if you have AI write without material. If you give it your own facts, your products and your wording rules, and add one piece of firsthand information before publishing, the text will sound like your company.
Is it better to publish automatically or after approval?
We recommend publishing after approval at first. Once you see how the AI writes and the amount you fix has dropped as you add to the material, a natural next step is to switch regular posts at fixed times to auto mode.
7 দিন বিনামূল্যে ব্যবহার করে দেখুন। ক্রেডিট কার্ডের প্রয়োজন নেই।
বিনামূল্যে শুরু করুন
