The competitors you're not watching. AI already is.

When a brand sets up to measure its AI visibility, the first thing it does is name its competitors — the three or four rivals it wants to be measured against. That list is a statement of belief: these are the firms we're up against. It is also, almost always, wrong at the edges. The engines don't read your competitor list. They build their own, from the whole web, every time a buyer asks — and the names they put next to you are frequently ones you never thought to track.

That gap has a cost. If a rival is surfacing in your answers and it isn't on your list, it is invisible to your measurement even as it is visible to your buyers. So the instrument does something your tracked list can't: on every Pro and Scale run, a language-model pass reads each stored answer and captures every commercial brand named in it — including the ones you've never heard of. When an untracked brand shows up often enough, it's flagged as a discovery candidate. We call the card "Also in your answers." It is the list you didn't make.

AIV Index research — the SA AI Visibility Index, Edition 1 — 1,296 grounded answers across South African property, law and accounting, measured June–July 2026 with open-capture extraction of every commercial brand named, not only tracked ones.

Your competitor set is a hypothesis. The engines don't share it.

A tracked competitor list is built from how a market looks from the inside — the firms you pitch against, lose deals to, watch on LinkedIn. It encodes your assumptions about the category. The trouble is that AI answers are built from how the market looks to everyone else: the directories, guides, reviews and comparison pages the engines read. Those sources don't rank brands by boardroom reputation. They rank them by who gets written about, and that is a different list.

So the brands the engines name alongside you split into two groups: the ones you expected, and the ones you didn't. The second group is the whole point. A rival you already track tells you nothing new. A rival you don't track — one the engines are quietly recommending in the same breath as you — is a blind spot with a name.

You can only manage the competitors you can see. The engines see more of them than you do.

In Edition 1, the unexpected names weren't a footnote — they were the story

The clearest case was accounting. Ask the engines who should handle a small business's books or tax, and the Big 4 don't top the answer. BDO is named in 28% of accounting answers, ahead of Forvis Mazars (20%) and PKF (16%); Deloitte and PwC sit mid-table, and EY places ninth of the twelve firms we tracked. A firm building its competitor list from the profession's pecking order would have watched the wrong names. And the single most telling entry isn't an accounting firm at all: Xero, a software platform, is named more often than KPMG. No audit firm would have put a bookkeeping tool on its competitor list. The engines put it at the top of the answer.

Law told the same story in a different key. In a field where no firm dominates the head-count, a consumer-facing practice that few corporate firms would have thought to track surfaced as often as the biggest corporate names — because the sources answering everyday legal questions describe it, and describe the giants less. On a tracked list built from the corporate rivalry, that firm simply wouldn't appear. In the engines' answers, it was unavoidable.

AIV Index research — Edition 1 accounting league table (432 grounded answers); brand mentions captured open-set, including non-firm platforms.

A rival you don't track is a rival you can't answer

Discovery isn't a vanity metric. Each unexpected name is a decision waiting to be made. Sometimes the right move is to add the brand to your tracked set so the next run measures your share against it properly — you can't close a gap you aren't scoring. Sometimes it's to study why the engines reach for it: which sources describe it, what those pages say, and whether the same sources could describe you. And sometimes it's simply to know, before a prospect does, that the assistant is putting a name you'd dismissed right next to yours.

What you cannot do is act on a competitor you never see. A measurement that only ever compares you to the rivals you pre-named will confirm your assumptions and hide everything outside them. Open capture is the correction: it lets the answer tell you who you're competing with, instead of you telling the answer.

The competitor that costs you the deal is rarely the one you were watching.

Like everything on the instrument, it's engine by engine — a brand can crowd your answers on Perplexity and be absent on ChatGPT — so discovery is read per engine, and adding a discovered rival to your set takes one click.

What to actually do about it

1. Read "Also in your answers" every run. Treat each unexpected name as a question, not noise: why is the engine naming it next to me?
2. Add the ones that recur. A brand surfacing run after run belongs in your tracked set — you can't measure your share against a rival you aren't scoring.
3. Trace the source, not just the name. The pages describing a discovered rival are the pages you're likely missing from. That's the lever.
4. Don't trust a competitor list you wrote alone. The market you pitch against and the market the engines describe are not the same list. Measure both.

Competitor discovery is a Pro-plan reading on AIV Index (Pro and Scale), captured on every run across all four engines. Which rivals the engines are putting next to your brand — expected and unexpected — is specific to you, and you can measure it directly. The published SA AI Visibility Index shows the same open-capture method run across whole industries.

A sibling piece covers the other half of the reading: once you're named, being named isn't being recommended.

Sources: AIV Index research — the SA AI Visibility Index, Edition 1 (1,296 grounded answers across South African property, law and accounting; read it here). AI answers shift over time; this is a point-in-time read of patterns we measured, not a guarantee.