How South African medical-aid schemes can take advantage of AI search
Local scheme ecosystem entrenched; international medical-comparison sites still meaningful.
SA-AEO-Bench v1.2 — pre-registered on OSF before data collection. 19,020 successful AI responses across GPT-5, Claude Sonnet 4.5 and Gemini 2.5 Pro. 232,254 cited URLs classified against a published source taxonomy. 100 South African brands, 12 industries. Re-run quarterly. Methodology and analysis code public at osf.io/w4az2.
This page is one industry’s view of that dataset. The per-brand cut is in the Scorecard, the full per-industry tables in the Q2 Index Report.
The pattern.
AI engines treat SA medical aid as a Discovery category. Ask ChatGPT for the best medical aid in South Africa and Discovery Health is named first by reflex. Bonitas and Momentum Health usually fill the second tier. Everyone else — Medihelp, Fedhealth, Bestmed, Profmed, GEMS, Polmed — shows up only when the question is about something specific. Discovery does what Discovery does, which is generate enough Vitality and product coverage every quarter to keep the engines anchored on it. The rest of the field shares a much smaller editorial pie.
Segment queries are where the field breaks open. Best medical aid for families. Best for students. Best for chronic conditions. Best for pensioners. Best for the self-employed. Each one pulls a different shortlist, because the SA consumer press has trained the engines to associate specific schemes with specific members. Profmed gets named for professionals. GEMS gets named for public-sector employees. Bestmed turns up on the budget question. A scheme that has never been editorially linked to a segment does not show up on the segment query — even when the underlying plan is competitive on paper.
Hellopeter is the SA medical-aid story that scares CFOs once they read a thread for the first time. Claim disputes anchor harder than any other complaint type, because they represent the exact moment a family decided their scheme had let them down. Gemini reads Hellopeter heavily. ChatGPT and Claude largely do not. That single asymmetry is the biggest per-engine dynamic in the category, and it matters because Gemini powers Google AI Overviews. A scheme with quiet ChatGPT exposure can still be losing consideration in the answer most members actually see. The per-engine treatment is in the SA AI Visibility Index Report. The picture for your scheme is in the free Scorecard.
Where AI gets its info.
Three to five SA sites carry most of the weight in medical aid. The names matter. The counts are reserved for the Index Report.
BusinessTech is the SA reference for medical-aid roundups, contribution-cycle reporting and plan-comparison pieces. Every other consumer site ends up citing it. A scheme that turns up in a BusinessTech roundup lifts across all four engines. A scheme quiet here starts every answer behind the schemes that are not.
MyBroadband owns the digital-health-product layer — Vitality changes, app launches, telemedicine features. The coverage trains the engines to associate "modern medical aid" with the schemes that show up there. Schemes without a digital story to tell forfeit that framing almost completely.
When the question is about Council for Medical Schemes rulings, NHI commentary or contribution-increase justification, AI reaches for Moneyweb. A scheme quoted on Moneyweb anchors more confidently in Claude than a scheme quoted anywhere else.
The Q2 Index Report names every SA domain we measured for medical aid — the full source taxonomy classified by role (default authority, freshness layer, complaint signal, long-tail specialist) with per-engine citation share for each one. The long-tail is where the cheap wins live. Download the Q2 Index →
For the source map specific to your brand — which domains currently shape AI’s answer about you — run the free Scorecard.
How the engines differ.
Each AI reaches for medical aid sources differently. One strategy almost never works across all four. The per-engine picture for your brand is in the Scorecard.
ChatGPT names Discovery Health by reflex. Bonitas and Momentum Health fill the second tier. It hedges on GEMS, Polmed, Profmed and the smaller schemes unless the question is explicitly about their segment. Best engine for "biggest medical aid" intent. Discounts Hellopeter almost entirely.
Claude leans on Moneyweb and BusinessTech commentary harder than the other engines. It is the most likely engine to flag a PMB nuance, a late-joiner penalty or a waiting period in the same answer it recommends a scheme. Best for brand-positioning questions. Weakest for "cheapest plan" intent.
Gemini pulls Hellopeter claim-dispute threads, Reddit and MyBroadband forums that ChatGPT and Claude ignore. The most dangerous engine for a scheme with active complaint volume, and the engine that powers Google AI Overviews — which is what most SA families actually see when they search.
Contribution-increase announcements, plan launches, CMS rulings — Perplexity catches them first. A scheme that fed the SA consumer press in the last fortnight reads as the current authority. A scheme that has gone quiet falls out fastest, especially through the October-to-January contribution cycle.
Two ways medical aid brands play this.
Brands in this category split into two camps. Which camp you are in tells you what AI-visibility risks you actually carry.
The category default
Discovery Health. The scheme AI engines treat as the safe answer on almost every question, with Bonitas and Momentum Health occupying a softer version of the same position. The advantage is reflex inclusion in nearly every recommendation set. The risk is reputational gravity — a single Hellopeter wave, a single CMS finding, a single contribution-hike backlash shows up faster and louder in Gemini answers than the brand team can usually respond to. Big schemes pay for their visibility in big news cycles.
The segment schemes
Profmed for professionals. GEMS for public-sector. Polmed for the service. Bestmed and Fedhealth on the value and self-employed tiers. These schemes have trained the engines to associate them with a specific kind of member through years of consistent editorial framing. They win decisively on the segment query and lose almost completely on the general "best medical aid" question. The risk is segment drift. The day the SA consumer press stops linking the scheme to its segment, the segment-anchor decays, and there is no general answer to fall back on. Recovery is slow.
What changed this quarter.
Questions to take to your next meeting.
Each one is specific. Each one is answerable. We answer them for your brand in the Scorecard and for the category in the Q2 Index Report.
- 01When ChatGPT recommends "best medical aid in South Africa", is your scheme in the answer or in the qualifying paragraph nobody reads?
- 02Which Hellopeter claim-dispute thread is currently shaping Gemini's answer about your scheme? Have you ever read it?
- 03For your closest competitor in the segment you serve: which engine treats them most generously, and what coverage explains the lift?
- 04How wide is the gap between your best AI engine and your worst? For most SA schemes it is wider than they expect.
The Q2 Index Report includes the full diagnostic question set for medical aid — every meeting-ready question with the answer cross-referenced to the per-engine measurement. Plus the SA brands currently ahead, behind, or improving fastest on each one. Download the Q2 Index →
For the answer specific to your brand — where you sit on each question, on each engine — run the free Scorecard.
What to do about it.
Three categories of move lift medical aid AI visibility. Which specific tactic is right for your brand depends on where you stand today — that conversation lives in the 30-minute walkthrough.
Audit your placement on the five SA medical-aid sites.
BusinessTech, MyBroadband, Moneyweb, IOL and Hellopeter are the five defaults. A scheme on the first four and quiet on the fifth reads very differently to AI than one with active claim-dispute volume. The Scorecard names where you currently show up and which placements have decayed since last quarter.
Map your segment-query coverage before defending the general question.
Segment queries — family, student, pensioner, chronic, self-employed — are where smaller schemes can win without out-spending Discovery. A scheme that owns two or three segments cleanly beats a scheme spread thinly across all of them. The Index Report quantifies the segment-anchor patterns across the sector.
Watch the Gemini-only signal you cannot see.
Hellopeter and the forums move Gemini, and Gemini powers Google AI Overviews. A scheme with quiet ChatGPT exposure can still be losing consideration in the answer most members actually see. The Q2 Index Report quantifies the gap. The Scorecard names the specific threads currently moving Gemini's answer about you.
Brand-level scorecards for medical aid land in the next Index Report. Reserve your copy →
Common questions,
answered straight.
Why does Discovery Health dominate almost every AI answer about SA medical aid?+
Discovery generates more SA consumer-press coverage per quarter than any other scheme — Vitality changes, app updates, product launches, app-store reviews — and the engines read that volume as authority signal. ChatGPT, Claude and Perplexity all anchor on it by reflex. Gemini still elevates Discovery but is more willing to name alternatives on segment queries because of its weight on forum and Hellopeter signal. The full per-engine breakdown is in the Q2 SA AI Visibility Index Report.
My scheme is on BusinessTech but not on Moneyweb. Does that actually matter?+
Yes — and it matters differently per engine. The two sites serve different intents: BusinessTech is the consideration-and-comparison authority, Moneyweb is the policy-and-regulator authority. Missing from one means missing from a specific category of question rather than from search overall. The Scorecard names which questions you are currently winning and which you are losing because of that gap.
Are AI engines actually accurate on PMB and late-joiner-penalty rules for my scheme?+
They are accurate at the regulator level and frequently wrong at the scheme level. AI explains correctly that PMBs must be covered in full and that late-joiner penalties exist. It regularly mis-attributes specific waiting periods, scheme-exclusions and benefit structures to the wrong scheme. The Index Report includes the categories of factual error we observe most often. The scheme-specific accuracy state is in the Scorecard.
How fast does AI-visibility data go stale in medical aid?+
Perplexity drops contribution-increase and plan-tier coverage older than roughly two months. Gemini decays next. ChatGPT and Claude hold longer but refresh aggressively through the annual contribution cycle from October to January. The full per-engine decay-half-life is in the Q2 Index Report. The freshness state for your specific scheme is in the Scorecard.
Where do you get this data?+
We run roughly four to six thousand structured prompts per quarter across ChatGPT, Claude, Gemini and Perplexity. Five replications per prompt. Latin-square order-reversal to control for position bias. Methodology pre-registered on OSF. See the full methodology for details.
Keep going.
Best medical aids in South Africa, AI-ranked
How the engines actually rank SA schemes — and where they disagree.
How SA medical-aid companies improve AI search visibility
The categories of move that lift scheme visibility — without giving away the playbook.
Methodology — SA AEO Bench
Pre-registered measurement design and source taxonomy.
Medical aid is the category.
You are the question.
The free Scorecard runs the same medical aid measurement against your specific brand. You get the per-engine gap, the SA sites currently shaping your AI answer, and the three moves that matter most for you.
- Dataset: sa-aeo-bench-v1
- Snapshot: 2026-05-19 · 17 days old · budget 270 days
- Pre-registration: osf.io/w4az2
- 14,826 responses · 188,877 citations · 100 brands · 10 industries
- POPIA-compliant · Information Officer registered
- Data residency: SA + EU only
- Methodology: pre-registered before data collection · audit trail per record
- Models measured: OpenAI GPT-5 · Anthropic Claude Sonnet 4.5 · Google Gemini 2.5 Pro