Webinar: AI Guardrails That Help Your Employees Post With Confidence

The program is not short of people, it is short of confidence

Koen opens with the number that shaped Ambassify's last year of product work. Looking across customer data, only around 10% of people in an organization are ever inclined to post on LinkedIn, and that share falls as headcount grows.

For some companies, ten highly active advocates out of a hundred is fine. For most, it is the reason the program quietly plateaus a year in.

"The program is not short of people. It is short of confidence."

The 90% who stay silent are rarely opposed to sharing. Ask them directly and you hear three things:

• "I don't know what to write."
• "What if I say something wrong?"
• "Nobody ever told me what I can and cannot say."

Koen's point is that these are three different problems with three different answers. Skills answer the first. Confidence-building answers the second. Guardrails answer the third. None of them are fixed by adding more features to a sharing tool.

He makes the same point about himself. As CEO he can post whatever he wants, and Ambassify's own system still catches things he should not say publicly, unreleased product features or competitive intel. Every company has those lines. Most never write them down anywhere an employee will read them at the moment they are typing.

AI-assisted, not AI-automated

The second half of Koen's opening is a deliberate position on AI.

The common setup today is AI writes the post, the employee clicks share. It works, it saves time, and for plenty of people it is the first step into sharing at all. Ambassify keeps those features. But Koen is blunt about the limits: the employee learns nothing from resharing a caption someone else generated, and the output starts to sound identical across every profile.

"The fear is that everyone starts to sound the same."

That matters more as LinkedIn fills up with AI slop. The authentic voices in a company are often not in marketing or HR. They are in finance and engineering, and those are exactly the people least confident about posting in their own words.

The direction Koen sets out is the opposite of autopilot:

"The employee writes, and the AI reviews."

Your policy, your brand guidelines and your social media rules are applied as guidance while the person is typing, so they learn from the experience instead of outsourcing it.

Layer one, company tone of voice

Kateryna takes over with the four layers, three live today and one in beta.

Tone of voice is a prompt the admin writes so the platform sounds like the company. It steers internal communication toward your ambassadors, and the more specific the instruction, the better the output.

Example instructions from the session:

• Speak to one team member, not to an audience as "we, the company"
• Avoid corporate language and specific jargon
• Use short sentences
• Keep it light and enthusiastic
• Do not overload with emojis

Her advice is to start simple, then adjust once you see what it produces.

Layer two, share message instructions

A separate prompt that applies specifically to captions, both the ones an admin writes and the ones a member gets from Improve with AI.

The useful part is that the two layers can pull in different directions on purpose. Your campaign copy can be informal while the captions that go out to LinkedIn stay formal. Admins and advocates save time, and the person posting stays responsible for authenticity.

In practice: an admin drafts a caption for a campaign and clicks generate with AI. The member sees that caption as a predefined message, can edit it, can run Improve with AI on their version, and then shares it.

Layer three, member tone of voice

Every member can set their own tone in their profile or at registration: friendly and casual, formal and polished, bold and energetic, or direct and clear.

This sits on top of the company tone of voice rather than replacing it. Your instructions apply first, their preference layers over. It is a fixed pick rather than a free text field, so members cannot write their own prompt, which keeps the output inside your guidelines.

It applies to personal posting and to editing share message captions. The result is more varied messaging across the community, which is the practical defense against everyone sounding the same.

Improve with AI is a second pair of eyes, not a ghostwriter

Kateryna is precise about what the live feature does. It does not write from scratch. It enhances what the person already wrote.

"It's a second pair of eyes. It's not a ghostwriter."

You find it in the personal post builder and in the edit share message screen. It reads both the company tone of voice and the member's tone of voice, and it gives the person a bit more confidence while keeping the messaging on brand.

Plain language beats policy language

Ambassify launched four skill modules: boosting content, sharing content, editing the share message, and creating a personal post.

Two of them are where admins have real control. In the edit share message and personal post skills you can configure posting guidelines per language, which is how you get your policy in front of people at the moment it is relevant.

"Most of your members do get a policy. It's several pages, and of course they forget about it."

The personal post skill shows what a strong post looks like, suggests what to write about, includes an editable example post, and ends with an interactive exercise. Alongside the composer, members see writing tips you configure in account settings: which hashtags to use, which tone to keep, what is worth thinking about before posting.

Personal posting is available on Professional and Enterprise plans. If you want to test the skill, contact your Customer Success Manager and it can be enabled on your account.

Layer four, AI Guardrails Share Review (beta)

Laurens closes with what the team is building now.

The concept is AI as a coach. Rather than checking a post after it goes out, the system checks it while it is being written and gives feedback in a panel next to the composer. His example in the session was a post reading "Ben is the worst employee ever," which returned one blocking issue and two suggestions.

Feedback comes back at two severity levels:

• Soft, a suggestion. You used a term we would rather avoid, here is why.
• Hard, a violation. The post cannot be published as written.

The screens shown are mockups, so the exact look can still change.

The six checks

1. Rules from your own policy. The most important layer, because it is specific to your company. Ambassify distills rules from your documents and you fine-tune them together. This is where you set what is encouraged and what is off limits.
2. PII detection. Personal identifiable information, on by default, and you can disable it.
3. Banned terms. Exact words and word groups. Insults, but also product names you want kept out of public posts because they cause confusion.
4. Optimization tips. Not company specific. Best practice from LinkedIn experts: start with a strong hook, structure the post, keep the length right. If someone writes one solid block of text, the tip suggests paragraph breaks. People learn while doing.
5. Brand voice check. Takes the share message instructions into account, so the layers work together rather than contradicting each other.
6. AI judgment. A trained model that flags slurs, threats, hate speech and insults. Built in, enabled by default, and you can turn it off.

Every check can be enabled or disabled. You keep full control over what the system looks for.

How setup and testing work in the pilot

The process Laurens described is deliberately light on your side:

1. Send Ambassify your documents. A social media policy, a charter, or house rules you wrote for your ambassadors.
2. Ambassify turns them into rules, each with a description and a severity level.
3. You fine-tune the rules together.
4. Ambassify runs the rules against posts your advocates have already published, so you can see what feedback the model would have given on real content.
5. As community manager, you give each piece of feedback a thumbs up or thumbs down. You can add a score out of five and written remarks if you want to go further.

The thumbs alone are enough to train the model on your account. Ambassify does most of the preparation work.

What comes next

The roadmap Laurens shared:

• Now: pilot testing with a small group of customers. More pilot customers are welcome.
• Phase 2: review available in the admin panel, so admins can quality check the captions and share messages they prepare before they go out.
• Phase 3: released to advocates, starting with personal posting.
• Phase 4: extended to share campaigns, so advocates get the same feedback when editing a share message you provided.

Do you share data with AI processors?

The question Ambassify gets most often, raised again in the chat.

No. All AI processors are based in the EU, not in the US. Ambassify is GDPR compliant.

Want to join the pilot?

If AI Guardrails Share Review fits a problem you have, contact your Customer Success Manager and Ambassify will set it up with you. For questions about the vision behind it, email Koen directly at koen@ambassify.com.

---