Which South African ISP Do AI Agents Recommend? Telkom vs Afrihost vs MWEB vs Webafrica
We ran an agent experience audit of Telkom, Afrihost, MWEB, and Webafrica in South Africa. Here's which one AI agents recommend, and why the AI's favorite pick doesn't always match the human reviews.
Ask an agent which fibre provider to sign up with in South Africa, and it names Afrihost most of the time, citing a real MyBroadband ISP of the Year award. Check Afrihost's own Trustpilot and HelloPeter pages, though, and its real customer sentiment is as bad as Telkom's, which every test in this audit, including the AI itself, ranks as the worst option.
Across 20 test runs, here are the results in summary:
- Afrihost is the agent's most consistent pick. It gets named unprompted most often and wins 4 of 5 head-to-head prompts outright or jointly, citing a real MyBroadband ISP of the Year award. Its own Trustpilot and HelloPeter scores tell a very different story.
- Webafrica wins 1 of 5 head-to-heads, on price and flexibility, and has a strong public review record to back it up.
- MWEB never comes out as the agent's outright top pick, consistently placing second or third, praised specifically for phone support.
- Telkom finishes last on Discovery and Sentiment. Agents rank it last in all 5 head-to-head prompts, 3 of them with explicit "avoid" language, and its own Trustpilot score is the lowest of the four.
Telkom isn't on the podium because it finished last, by a wide margin, on every test in this audit. The rank combines each ISP's Discovery and Sentiment results below, weighted equally.
Two runs give a general sense of what an agent does with a question like this, before we get into the detail.
Asked to pick exactly one best-value ISP of the four, and told not to hedge, the agent commits to Webafrica.
- It runs two web searches first, comparing all four companies and then checking prices specifically. The answer is built on what those searches return.
- The verdict is a single word, "Webafrica.", followed by three reasons: lowest prices at every speed tier, no contract lock-in, and unshaped lines on all the major networks.
- Then it does something more interesting. It names Afrihost's MyBroadband ISP of the Year award and sets it aside, "that's a support/reputation win, not a value one."
Given a question specific enough to force a choice, the agent will look past the award. Most questions aren't that specific.
Telkom was never recommended as a pick in any of the 5 Discovery prompts we ran, alone among the four ISPs in never getting named as a choice unprompted. Asked how to sign up for home fibre, the agent named Webafrica, Afrihost, Cool Ideas, and RSAWEB as the ISPs to compare, and left the country's largest ISP out of its own category.
Telkom's real MyBroadband Q1 2026 satisfaction score and its 1.3/5 Trustpilot rating back up what the agent already concludes on its own. Named directly, it finishes last on every test in this audit, as the Sentiment section below shows.
What agent experience means here
Agent experience (AX) is how well an AI agent can find, understand, and recommend your product when someone asks it for help.
We evaluate that based on two things:
- Discovery: does the agent name this company unprompted when potential customers don't ask about it by name?
- Sentiment: once all four companies are named directly, which one does the agent prefer, and why?
Yes, model output is non-deterministic and your own context can shift what an agent says. That's why we ran each prompt shape across both web-search and no-search passes. The pattern that held up across those runs is what this article reports.
Do agents know these companies exist?
Before someone asks an agent to compare specific ISPs, they usually just describe what they need:
- "What's the best fibre internet provider in South Africa?"
- "Which ISP should I sign up with for home fibre in Cape Town?"
Whichever company the agent already knows about gets a free recommendation.
Asked the most generic version of that question, the agent won't commit to anyone:
What's the best fibre internet provider in South Africa?
The agent declines to pick a winner at all. It deflects to infrastructure, "it depends more on which network is available at your address than on the ISP brand," then lists Cool Ideas, RSAWEB, Afrihost, and MWEB generically without ranking them. Notably, none of the four is Telkom or Webafrica. It closes by asking for a suburb rather than committing.
Give it a specific problem to solve, though, and it commits:
I want to switch ISPs but I'm worried about getting stuck between departments who each blame the other. Which South African ISP has the smoothest, most self-service signup process?
- This run triggers three real web searches before answering, then commits: "Afrihost is the strongest pick for this specific worry."
Its reasoning leans on exactly the material the agent can find and cite:
- Afrihost wins on its MyBroadband ranking, 72.7% in the Q1 2026 satisfaction survey and a fourth year running as ISP of the Year, ahead of MWEB at 72.6% and Webafrica at 70.1%.
- Webafrica is demoted to "also worth considering" despite the agent crediting it with the "best raw customer-experience score (4.4/5 on Trustpilot, ~17,000 reviews, more than every other major SA ISP combined)."
That gap is the article in miniature. The agent picks the company with the better award record over the one with the better customer reviews, and it says so in the same breath.
The same pattern decides who gets left out. Asked how to sign up for fibre at all, the agent named Webafrica, Afrihost, Cool Ideas, and RSAWEB as the ISPs to compare, and left out Telkom, the country's largest. Openserve, Telkom's own fibre network, appears in the same answer as infrastructure to check coverage on, so the agent knows the company exists. It just doesn't think of Telkom as something you buy from. If an ISP isn't in the source material an agent finds, it can't end up in the answer.
This is the pattern that held up across our testing:
- Afrihost and Webafrica are the common defaults once the agent searches
- MWEB only comes up as a secondary, "slightly pricier" alternative
- Telkom is never recommended in any Discovery prompt we ran
Trends in the data
We ran this test 10 times, across web-enabled and no-web configurations, with prompts like:
- "What's the best fibre provider in South Africa?"
- "I want to sign up for fibre without being passed between departments. Who makes that easiest?"
- "What's the cheapest reliable fibre in South Africa?"
Here's how often each company got named:
- Afrihost: named in 4 of 5 web-enabled runs, across both prompt framings, and the outright pick on the department-shuffle prompt
- Webafrica: named in both of the runs that triggered a real web search, usually as the value or review-record option rather than the lead pick
- MWEB: named in 2 of 5, never as the lead recommendation
- Telkom: never named as a pick, in any run
What this means for each company
Getting found unprompted comes down to what's already been written about a company.
Webafrica is the closest thing to a default recommendation in this market.
- Named in both of the Discovery prompts that triggered a real web search, credited with the "best raw customer-experience score" of any major SA ISP
- Cited for its Trustpilot review volume, around 17,500 reviews at 3.9-4.4/5, more than every other major SA ISP combined
- Also named as one of the two cheapest options, at roughly R299 a month
- Keep the review volume and pricing pages current, since that's the source material agents are citing
Afrihost gets named across both prompt framings once the agent searches.
- Appeared in 4 of 5 web-enabled Discovery runs, more consistently than any ISP except Webafrica
- Agents lean on MyBroadband's ISP of the Year award, won four years running, and its 72.7% Q1 2026 satisfaction score, the highest of the four
- The outright pick on the department-shuffle prompt, chosen for its ClientZone self-service dashboard and its MyBroadband ranking
- Keep the awards and survey results agents cite current, since the visibility rests on that material rather than on the product itself
MWEB gets mentioned, but only ever as an also-ran.
- Named in 2 of 5 web-enabled Discovery runs, and never as the lead recommendation
- When it does appear, it's framed as the "slightly pricier" alternative to whichever ISP the agent named first
- Absent entirely from the two prompts where the agent searched and committed to a specific pick
- Get into the comparison articles and community threads where Afrihost and Webafrica already appear, rather than waiting on training data to shift
Telkom is never a pick, despite being the biggest fixed-line operator in the country.
- Never recommended as a choice in any of the 5 Discovery prompts, in either mode
- Appears only inside generic explanations of how the ISP and network-operator split works, alongside a list of names
- Left out entirely when the agent named the ISPs to compare for a fibre signup, including much smaller players, while listing Telkom's own Openserve network as infrastructure in the same answer
- Publish comparison-friendly content (pricing, coverage, packages) so Telkom appears in the round-ups agents already pull the other three from
Which company do agents prefer?
This section is about what happens once a buyer names all four companies and asks the agent to choose. A company can be named first by default and still lose a direct comparison.
What's the best choice between Telkom, Afrihost, MWEB, and Webafrica for home fibre internet in South Africa?
The agent names Afrihost "the best overall choice," citing MyBroadband's ISP of the Year award for the fourth consecutive year and the highest satisfaction score of the four at 72.7%. MWEB follows at 72.6% and Webafrica at 70.1%. Telkom, it notes, "didn't rank in the top tier of these surveys."
Webafrica gets credited in the same breakdown with the "best customer experience on Trustpilot (4.4/5 from ~17,000 reviews)," and still doesn't win. The award record beats the review record again.
Ask directly about Telkom, and the agent doesn't hedge:
Telkom reviews? Should I use it?
"Telkom's reviews are notably poor, I'd avoid it in favor of Afrihost, MWEB, or Webafrica." It backs that with a 1.3/5 Trustpilot score across 417 reviews, 2,691 ComplaintsBoard complaints with only 1% resolved, and recurring themes of installation delays, billing disputes, and chatbot-only support. Asked whether to use it, the answer is "Not recommended based on this evidence."
Trends in the data
We ran this test 10 times, across the same two configurations, in two prompt shapes. Short, direct four-way comparisons:
- "Telkom, Afrihost, MWEB, or Webafrica. Which one should I use for fibre?"
And longer, problem-framed questions:
- "I want to sign up for fibre entirely online without having to phone a call centre or deal with being passed between departments. Between Telkom, Afrihost, MWEB, and Webafrica, which one lets me do that?"
Here's who came out on top:
- Afrihost: the most consistent winner, top or joint-top pick in 4 of 5 prompts, called the "best overall choice," the "safest overall pick," and the "strongest fit" across different runs
- Webafrica: top or joint-top in 2 of 5, tied with Afrihost on the signup-process prompt and the outright winner on the one that forced a single unhedged answer
- MWEB: never picked outright, placing second or third every time, praised specifically for phone-support responsiveness
- Telkom: ranked last in all 5, with explicit "avoid" language in 3 of them and "riskiest, based on complaint patterns" on the signup-process prompt
Afrihost wins the AI's preference and a real MyBroadband ISP of the Year award, four years running. Its own public review pages say something else entirely:
- MyBroadband Q1 2026 satisfaction: 72.7%, the highest of the four
- Trustpilot: ~1.3-1.4/5 ("Bad"), from around 280 reviews
- HelloPeter: a 1.8 TrustIndex, across 913 reviews
That puts Afrihost in the same range as Telkom, the ISP the agents themselves tell buyers to avoid.
Afrihost's "safest overall pick" status rests on the metric easiest for an AI to cite, published awards and survey percentages, and collapses on the metric a buyer hits first when they search the name themselves.
What this means for each company
Winning a direct comparison comes down to specific, checkable claims about prices, service, and reputation.
Afrihost wins the head-to-head, on evidence a buyer may not agree with.
- Top or joint-top pick in 4 of 5 Sentiment prompts, called the "safest overall pick" and the "strongest fit"
- Agents justify it with the MyBroadband award and a 72.7% satisfaction score, the highest of the four
- Its own Trustpilot page reads ~1.3-1.4/5 ("Bad") from around 280 reviews, and HelloPeter puts it near 1.8, both in Telkom's range
- Close the gap between the awards story and what Trustpilot and HelloPeter show, since that's the next thing anyone doing their own due diligence will hit
Webafrica takes the value comparison outright, but not the open-ended question.
- Top or joint-top pick in 2 of 5 prompts, including the single prompt that forced an unhedged answer
- Won that prompt on cheapest entry pricing with no lock-in, and on review volume
- Loses the broader "which should I use" framing to Afrihost, which agents treat as the safer default
- Publish clearer, more citable pricing and review data so the value case carries beyond the cheapest-option question into the general one
MWEB is respected in every comparison and chosen in none of them.
- Never named outright first, placing second or third in all 5 prompts
- Singled out for phone-support responsiveness, its only attribute agents consistently credit
- That strength is also used against it. On the self-service prompt, agents marked MWEB down for leaning on phone support
- Publish specific, citable detail on the parts of the service that don't need a phone call, since the support-quality case is already landing
Telkom loses every comparison, and the agents give reasons a buyer can check.
- Ranked last in 5 of 5 Sentiment prompts, with explicit "avoid" language in 3 of them
- Called the "riskiest" option on the prompt about signing up without a call centre, "based on complaint patterns"
- Backed by real data. A 1.3/5 Trustpilot score from ~430 reviews (93% one-star), 1.15/5 on HelloPeter across 3,318 reviews, and the lowest MyBroadband Q1 2026 satisfaction score of the four at 66.9%
- Fix the underlying service and support experience before the messaging, since agents are citing real satisfaction data rather than vague reputation
What this means for this market
Afrihost leads on both Discovery and Sentiment, but its own review pages contradict the AI's confidence in it. That gap between what agents repeat and what a buyer finds on their own is where the opportunity sits for whichever ISP moves on it first.
Afrihost: the AI's confidence is running ahead of the reviews
Winning both Discovery and Sentiment is valuable, but a real, checkable gap between what the AI says and what Afrihost's own Trustpilot and HelloPeter pages show is exactly the kind of thing a buyer, or a more thorough agent, finds the moment they look past the headline recommendation.
Webafrica: the product case and the review case both hold up
Webafrica wins on price when compared directly, and its public review record backs that up. The opportunity is making that pairing more visible earlier in the funnel, before a buyer gets to the head-to-head question.
MWEB and Telkom: close the visibility gap, and for Telkom, fix the underlying service too
Neither ISP is the AI's outright pick. For MWEB, that's a visibility problem on a service that already gets praised for support once it's compared directly. For Telkom, every test in this audit points the same direction, worst AI sentiment and the lowest real customer-satisfaction score of the four, so the fix starts with the underlying service before the messaging.
How to improve your product's agent experience
More and more buyers are letting an agent do the first round of research before they ever visit a company's website. If the agent doesn't know a company exists, or repeats a reputation the buyer can disprove in one click, that company is out before it gets a chance to sell anything.
A few things any ISP, or any technical product, can do to show up correctly when an agent is asked about it:
- Read your own review pages before an agent does. Agents cite Trustpilot and HelloPeter directly. Afrihost is the AI's most consistent pick while its own Trustpilot page reads as badly as the ISP every agent tells buyers to avoid, and that gap survives exactly as long as it takes a buyer to click the link.
- Get into the comparison and round-up content agents search for. Discovery depends on being in the source material a search turns up. MWEB is a top-four ISP that went unnamed as a pick in every Discovery prompt, while smaller providers like Cool Ideas and RSAWEB came up repeatedly.
- Publish specific, current figures. The prompts that produced real recommendations were the ones about price and self-service signup, answered with cited numbers. Vague positioning gives an agent nothing to quote.
- Expect to be judged on how you make people buy. Agents ranked these ISPs partly on whether a customer can sign up without phoning anyone. MWEB gets marked down on that prompt for the phone-support quality it gets praised for elsewhere.
Want help understanding and fixing the AX of your own technical product or platform? At Ritza, our Engineering Writers work at agent speed with human-expert verification (no slop) to win at GTM.