060 279 5587 info@sitect.co.za 139 Davies Street, Doornfontein, Johannesburg, 2001 Gauteng, SA
Services / AI Consultation & Integration / AI Content & Marketing Automation
Brand-trained · Multi-channel · A/B tested

Marketing that writes itself. (Almost.)

A content factory trained on your brand voice — drafting blog posts, social, ad copy, email and product descriptions across every channel you ship on. With a human-in-the-loop approval flow so nothing weird leaves the building.

Live in 3 weeks Human approval gates A/B testable From R9,000

Four reasons your content calendar is empty by Wednesday.

SA marketing teams are smaller and more stretched than anywhere else we work. Every team we audit has the same four bottlenecks.

!

Writer-bottleneck

One person writes everything. They go on leave. The calendar dies. The blog hasn't published since March. You've been "about to hire" for 8 months.

1 in 4posts cancelled when writer's out
!

Brand voice drift

Three writers, three vibes. The LinkedIn post sounds like a sales pitch, the blog like a research paper, the IG caption like a fever dream. Your brand has multiple personalities.

62%of audited posts off-brand
!

Same email to everyone

Your CRM has 14,000 contacts. Your newsletter goes to all of them, same copy, same time, same offer. Personalisation = "Hi {{first_name}}".

2.1%avg click-through baseline
!

Zero experimentation

You write one version. You ship it. You hope. No A/B, no variant, no learning. Six months later you can't say what works.

0A/B tests run last quarter

Eight content types. One brand voice.

The same trained model writes for every surface — so a Friday LinkedIn post sounds like the Tuesday newsletter sounds like the Sunday ad. Cohesion at scale.

Long-form blog posts

SEO-tuned 800–1,500 word articles with H2/H3 structure, meta tags, internal links and an FAQ block — drafted from a brief or a topic.

Speed~6 min per post

Social posts

LinkedIn, X, Instagram, Facebook — each platform gets its native format (LI hooks, X threads, IG captions). Multiple variants per topic for A/B.

Speed30 sec each

Email campaigns

Newsletters, drip sequences, win-back flows, promos. Segment-aware subject lines, body copy, CTAs. Multi-variant generation built-in.

Speed~3 min per email

Ad copy

Google Ads (RSA + Performance Max assets), Meta (5 headline + 5 body variants), LinkedIn Ads. Word-count-aware, A/B-ready, brand-checked.

Speed~2 min per set

Product descriptions

Bulk-write 500–5,000 SKU descriptions for your e-com store. Shopify, Woo, Magento — pulled, written, pushed back as drafts.

Speed~10 sec per SKU

SEO meta & schema

Title tags, meta descriptions, OG cards, schema.org JSON-LD for every page. Variant-tested for CTR in Search Console.

Speed~1 min per page

SMS & WhatsApp

Short-form, opt-in respectful, link-shortened. Birthday, abandoned-cart, payment-reminder, NPS flows. POPIA-compliant.

Speed~15 sec per msg

Landing-page variants

Hero copy, sub-heads, body sections, CTA buttons — multiple variants per audience segment. Plug into your A/B tool.

Speed~5 min per page
The brand voice trainer

Your voice. Fingerprinted.

Generic AI writes generic content. Sitect's content engine starts with a brand voice fingerprint — captured from your existing best writing, then tuned per channel. Every output checked against the fingerprint before it ships.

  • Voice axes — tone, formality, energy, irreverence, technicality scored on a 0–10 scale for your brand.
  • Lexicon library — your power words, your bans, your jargon, your "we say it like this".
  • Channel adaptations — formal on LinkedIn, punchy on X, warm on email — same brand, different gear.
  • Brand-score gate — every draft gets a score 0–10. Below threshold? Re-drafted automatically.
  • Continuous tuning — every approved piece feeds back into the fingerprint. The model gets sharper monthly.
  • POPIA-aware — opt-in respect, no manipulative urgency tactics, consent baked in.

15 channels, one calendar.

Drafted, approved, scheduled and published to every surface your customers are on. One source of truth — no copying between tools.

LinkedIn
posts · articles
X / Twitter
posts · threads
Instagram
captions · reels
Facebook
posts · groups
TikTok
scripts · captions
YouTube
titles · descriptions
Email
campaigns · drips
Blog / CMS
long-form · SEO
SMS
Clickatell · Twilio
Push notifs
web + mobile
Web copy
landing pages
Ads
Google · Meta · LI
Pinterest
pin titles · boards
Reddit
subreddit-tuned
WhatsApp
broadcasts · status

One brief. Five channel outputs.

This is a real example from a Sitect-built content engine for a Pretoria-based legal-tech startup. One paragraph of brief → 5 native, brand-consistent outputs in under 60 seconds.

→ Brief input

"Launch our new wills feature"

We just shipped automated will-drafting for South African families. R899 flat fee. Built with AI but every draft reviewed by our admitted attorney.

Key benefit: most people delay wills because lawyers are expensive. We make it cheap and dignified.

audience: working SA families tone: warm, candid CTA: try-it-free
LinkedIn post
187 words · 7 emoji-free
Most South Africans don't have a will.
Not because they don't want one. Because the average local attorney charges R3,500 just to write a basic one — and life gets in the way of "I'll sort it eventually."
We just shipped something we wish existed when our own families needed it…
X / Twitter thread
5 tweets · hook-first
Most SA adults don't have a will. 🧵
Reason isn't apathy — it's R3,500 to a lawyer.
We built something. R899. Drafted by AI, reviewed by our attorney. Dignified, fast, legal…
Email newsletter
subject + body · 4 subject variants
Subject: The R899 will you actually have time for.
Hi [first_name] — we just shipped something we've been quietly building for a year. A will. Drafted by AI. Reviewed by our attorney. R899 flat…
Instagram caption
128 chars · 3 hashtags
You don't have a will, do you?
We made it cost R899 instead of R3,500. Built with AI. Approved by our attorney. Link in bio.
#estateplanning #SAlegaltech #fintech
Google Ads (RSA)
5 headlines · 4 descriptions
Headlines: Wills for R899 · AI-drafted, lawyer-reviewed · Skip the R3,500 attorney
Descriptions: Done in 20 min. Backed by an SA-admitted attorney. POPIA-safe. Try free.

Brief in. Calendar out.

Five stages from a one-paragraph brief to scheduled, brand-checked posts across every channel — with humans-in-the-loop where it matters.

01 · BRIEF

Capture

Topic, audience, CTA.
  • Web form · Notion sync
  • Slack /post command
  • Strategy upload
02 · DRAFT

Generate

Channel-native outputs.
  • Brand voice tuned
  • Multi-variant
  • SEO + length aware
03 · SCORE

Brand check

Auto-validated.
  • Brand-score 0–10
  • Banned terms filter
  • Plagiarism check
04 · APPROVE

Human gate

Configurable per channel.
  • Slack inline approve
  • Edit-in-place
  • Re-draft button
05 · SHIP

Schedule + post

All your channels.
  • Buffer · Klaviyo · CMS
  • A/B variants
  • Performance loop

Six artefacts. All editable by your team.

The brand voice fingerprint, the prompts, the workflows — all stored in plain-text registries your team can edit without a developer.

01 · Brand voice

Voice fingerprint + lexicon

Trained on 50–100 of your existing best pieces. Tone axes, channel adaptations, always/never word lists. Lives in a JSON file your team can edit.

JSON + UI editor
02 · Content engine

Self-hosted draft API

The core engine. POST /generate with a brief and a channel, get back N variants with brand scores. Yours, in your infra.

Docker image
03 · Calendar dashboard

Plan, draft, approve in one UI

Web UI where your marketing team plans, drafts, reviews and approves. Slack notifications for pending approvals.

Next.js · self-host
04 · Publishing connectors

3–5 channels wired up

LinkedIn, Buffer / Hootsuite, Klaviyo / Mailchimp, your CMS, Google Ads — whichever 3–5 you ship to today. OAuth + token-vault wired.

OAuth + webhooks
05 · Variant + A/B engine

Multi-version everything

Every output is a set of variants — different hooks, different CTAs, different angles. Plug into your A/B tool, learn what works.

Variant comparison UI
06 · Performance loop

Wins feed back into the model

Post performance (opens, clicks, engagement) pulled back from your channel tools. Top-performing patterns reinforced in the next draft.

Weekly auto-tune

3 weeks. 4 phases. One brand voice fingerprinted.

The first piece of content ships in week 2. Week 3 is about refining the voice fingerprint based on real reactions.

01
Week 1 · 3 hrs

Voice fingerprint

You send 50–100 of your best existing content. We build the brand voice fingerprint, you sign off.

02
Week 2 · 1 hr

Wire channels

OAuth connect to your social, email and CMS tools. Test publish on each. Configure approval rules per channel.

03
Week 2 · 2 hrs

Pilot week

Run a real content week through the pipeline. Approve / edit / re-draft. Tune the prompts.

04
Week 3 · 1 hr

Go live + train

Full launch. 2hr training for your marketing team. 60 days of weekly tuning included.

The numbers behind a content engine done right.

Median results across our deployed content pipelines. Your numbers will be yours — these are the order-of-magnitude bracket.

Content velocity
Median lift in posts shipped per week vs the manual baseline.
94%
Brand consistency
Drafts passing the brand-score gate on first attempt.
+38%
Engagement lift
Median email + social engagement vs the previous quarter.
R28k
Monthly savings
Median freelance + agency cost displaced per month.

Pick the scope. We deliver in 3–4 weeks.

Build fee is fixed in writing. Variable cost = LLM tokens (~R0.50–R3 per generated piece) billed direct to your provider. No per-post platform fee.

Solo channel

Starter engine

R9,000
/project · ex 15% VAT
  • 1 channel (e.g. LinkedIn OR email)
  • Brand voice fingerprint
  • Up to 3 content types
  • Basic approval workflow
  • 30 days post-launch tuning
Choose Starter →
Enterprise

Content factory

R48,000+
scoped per engagement · ex VAT
  • All channels + custom connectors
  • Multi-brand voice support
  • Per-language tuning
  • SSO + role-based approval
  • Compliance review workflow
  • SLA + dedicated PM
  • 90 days hyper-care + QBR
Talk to us →

Common questions from marketing teams.

If yours isn't here, send a sample of your existing content — we'll come back with 3 brand-matched drafts in 48h. No sales call.

Will the AI content sound like us — or like every other AI-generated post?
It'll sound like you. The brand voice fingerprint is the difference. We train on 50–100 of your best existing posts, tagging tone, lexicon, structure. Every draft is scored against the fingerprint and re-drafted if it scores low. You'll get sample posts to compare before signing — if they don't sound like you, the engagement doesn't proceed.
What about plagiarism / duplicate content?
Every draft passes through a plagiarism checker (Copyscape API) and a duplicate-detection step against your own past content. Search-Console penalties for thin / spun content are a real risk — we engineer around them. SEO-content drafts include the AI-content disclosure recommended by Google Search guidelines.
Will my writers be replaced?
Honest answer: probably not replaced, but reshaped. Most clients keep their content lead in an editor / strategist role — running the brief, approving drafts, planning the calendar. The junior copywriter role usually shifts to "performance analyst" — measuring what's working and feeding learnings back into the engine. Teams typically ship 8× more content with the same headcount.
How does the brand voice fingerprint work technically?
It's a structured JSON file containing: (a) tone-axis scores (0–10 on formality, wit, technicality etc.), (b) lexicon arrays (always-use, never-use, brand-specific terms), (c) per-channel style overrides, (d) example pieces. The LLM prompt is constructed at runtime from this fingerprint. You can edit any of it in the dashboard or directly in Git.
Can it write in multiple languages?
Yes — English + isiZulu + Afrikaans are production-ready. Sesotho, Xhosa, Setswana are in beta. Each language gets its own fingerprint tuning since "casual" in English doesn't map 1:1 to "casual" in isiZulu. We're upfront when a language hits limitations.
What about Google's stance on AI content for SEO?
Google has confirmed they rank by quality, not by who wrote it. AI content done well — well-researched, original angle, real expertise — ranks fine. Spammy mass-produced content (which is what gives "AI SEO content" a bad rep) gets penalised regardless of who wrote it. We optimise for the former, with human editorial review on every long-form piece.
How do you handle facts & statistics?
By default the engine is forbidden from inventing statistics. If a brief contains stats we use them; otherwise drafts use qualitative language ("most", "many") instead of numbers. For verifiable facts (your product features, your pricing, your team) we wire up RAG against your knowledge base so the engine pulls directly from your truth.
What if a draft accidentally publishes something embarrassing?
Three safeguards: auto-validation against banned terms / phrases / topics; approval gates per channel (you can require human approval for everything, or only for posts above a brand-score threshold); kill switch in the dashboard pauses all scheduled posts in one click. The audit log shows every approval + edit.

Send us your brand.
We'll send back 20 posts.

Drop 10 of your best existing posts into our form. Within 48 hours we'll send back 20 fresh drafts in your voice — across 4 channels — plus a brand-voice fingerprint document you can keep, even if you don't hire us. No sales pitch.

You'll get back in 48h

  • 20 brand-tuned drafts
  • 4 channels (LI · X · email · ads)
  • Your brand voice fingerprint
  • Honest "is this worth it" assessment

More AI Consultation services

AI Strategy & Roadmap

Where does AI fit in your business?

AI Chatbot Integration

Smart chatbots that capture, qualify and convert.

AI Voice & Phone Agents

24/7 AI agents that handle calls.