Skip to main content
AIV Index← All research

Research · Recurring index

The SA AI Visibility Index.

When a South African buyer asks an AI assistant “who’s the best [industry] near me?”, a shortlist comes back — and most brands aren’t on it. The Index measures who is: grounded answers from ChatGPT, Claude, Gemini and Perplexity, scored per industry, with the method and every prompt published.

Edition 2 — October 2026 · Published 9 October 2026 · 3 verticals · 1 114 grounded answers · Baseline: Edition 1 — July 2026, restated

Edition 2 — October 2026 · Published 9 October 2026

The shortlists held. Two platforms fell.

Edition 2 asked the same 108 buyer questions of the same four engines on 25 September 2026 and compared every tracked brand with Edition 1, under rules published before the sweep. No tracked estate agency, law firm or accounting firm moved by 5 points or more on both sample rules (the second rule is explained below). Two platforms did: Property24 and Xero are named in fewer answers than in Edition 1.

Edition 1 to Edition 2: brands that moved 5 points or more on both sample rules, brands whose move depends on the rule, and the number that did not move.
VerticalMoved 5+ pointsDepends on the ruleNo change
PropertyProperty24 −5.7none13 of 14
Legalnonenone14 of 14
AccountingXero −7.1BDO South Africa −5.6, SNG Grant Thornton −5.0, Deloitte South Africa −7.810 of 14
The band
Fixed before the sweep: a share of mention moving by fewer than 5 percentage points is no change, and so is a swap in rank between brands fewer than 5 points apart. Changes are computed on the unrounded shares. Every comparison is against Edition 1 as restated on 3 September 2026.
The sample rule
Edition 2 counts an answer only if the engine searched the web for it. Our ChatGPT arm declined to search in 151 of its 324 answers and Claude in 31. Edition 1 could not see this, so it counted them. A comparison across that change can move for a reason that has nothing to do with the brands, so a move is reported only if it also holds with those answers put back, as Edition 1 counted them. ChatGPT’s and Claude’s own columns are not compared across editions at all.
The engines
The same four models were requested as in Edition 1. ChatGPT and Claude reported the same dated model on every Edition 2 answer (gpt-4o-mini-2024-07-18, claude-haiku-4-5-20251001). Edition 1 did not record the model the provider ran, so a change between the editions cannot be ruled out for any engine, and Gemini and Perplexity report only the name we asked for. No engine is marked not comparable, but the check could only have caught a change inside Edition 2, on ChatGPT or Claude.
When
The sweep was pre-registered for 15 September 2026; it ran on 25 September 2026, ten days late. Publication was pre-registered for 6 October 2026; it published on 9 October 2026, three days late. Edition 1 was measured in June (property) and July (legal, accounting).

What this cannot show: why anything moved. Between the editions the engines updated, sites changed and competitors published. The Index records what the engines said on two dates with one instrument, and the method note sets out every rule above with its counts.

Residential property & estate agencies

Measured 25 September 2026 (Edition 1: June 2026) · ChatGPT (gpt-4o-mini) · Claude (claude-haiku-4-5) · Gemini (gemini-2.5-flash) · Perplexity (sonar) · 36 prompts × 3 runs × 4 engines = 432 answers, 375 grounded (57 declined searches left out)

In September 2026, the South African estate agencies most often named by AI assistants were Pam Golding Properties (46% of answers) and Seeff (37%), in the same order as in June, with Rawson, Engel & Volkers, RE/MAX and Chas Everitt within four points of each other behind them. Measured across 375 grounded answers from ChatGPT, Claude, Gemini and Perplexity.

Residential property & estate agencies: share of AI-assistant answers naming each tracked brand, Edition 2 — October 2026, measured 25 September 2026, with the change from Edition 1 in percentage points.
#BrandShare of mentionEdition 1Change, pointsShortlist shareConviction
1Pam Golding Properties
46%
45%+0.955%97%
2Seeff Properties
37%
35%+1.838%93%
3Rawson Property Group
20%
20%+0.416%74%
4Engel & Volkers Southern Africa
18%
16%+1.428%88%
5RE/MAX of Southern Africa
17%
22%−4.514%91%
6Chas Everitt International
17%
20%−3.319%81%
7Lew Geffen Sotheby's International Realty
13%
13%−0.412%85%
8Jawitz Properties
12%
13%−1.29%84%
9Greeff Christie's International Real Estate
8%
5%+3.49%84%
10Dogon Group Properties
7%
5%+2.66%93%
11Harcourts South Africa
7%
5%+1.85%73%
12Just Property
7%
10%−2.86%89%
—Property24platform
10%
15%−5.7 down—36%
—Private Propertyplatform
6%
10%−4.0—50%

Share of mention = share of all 375 grounded answers naming the brand. Edition 1 = the restated Edition 1 share. Change = Edition 2 minus Edition 1, in percentage points; a move under 5 points is no change, and only moves of 5 or more are labelled. Shortlist share = the same measure on “best estate agents” questions only. Conviction = of the answers naming the brand, the share presenting it as a recommended choice.

Can the engines read the sites?

Residential property & estate agencies: whether each tracked brand’s website served the AI crawlers, by the Edition 1 method and by the search-crawler method new in Edition 2, probed 9 October 2026.
BrandEdition 1 methodSearch crawlers (new)
Pam Golding Propertiesvisiblevisible
Seeff Propertiesvisiblevisible
Rawson Property Groupvisiblevisible
Engel & Volkers Southern Africavisiblevisible
RE/MAX of Southern Africablocks GPTBot (403)visible
Chas Everitt Internationalvisiblevisible
Lew Geffen Sotheby's International Realtyvisiblevisible
Jawitz Propertiesvisiblevisible
Greeff Christie's International Real Estatevisiblevisible
Dogon Group Propertiesvisiblevisible
Harcourts South Africavisiblevisible
Just Propertyvisiblevisible
Property24visiblevisible
Private Propertyvisiblevisible

Edition 1 method = GPTBot, ClaudeBot and PerplexityBot, served or refused at the time of the fetch, kept so the two editions compare like for like. Search crawlers = OAI-SearchBot, Claude-SearchBot and PerplexityBot, the crawlers that feed each engine’s search answers, with robots.txt opt-outs counted and a normal-browser fetch as the control. Probed 9 October 2026.

The engine you ask changes the answer

Residential property & estate agencies: share of mention per AI engine for the most-named brands.
BrandChatGPTClaudeGeminiPerplexity
Pam Golding Properties26%54%54%41%
Seeff Properties20%41%42%37%
Rawson Property Group2%20%38%13%
Engel & Volkers Southern Africa10%15%11%32%
RE/MAX of Southern Africa2%16%32%11%
Chas Everitt International13%13%24%15%

The ChatGPT and Claude columns count only the answers where that engine searched the web: 61 and 98 of 108 here. Edition 1 could not tell which answers were searched, so these two columns are not compared across editions.

The top of the category did not move. Pam Golding Properties is named in 46% of answers, against 45% in Edition 1, and presented as a recommended choice in 97% of the answers that name it. Seeff holds second. Below that, four agencies sit within four points of each other, in a different order from Edition 1; a reshuffle that close is not movement under the pre-registered rule.

Property24 is the one property move past the band on both sample rules: named in 10% of answers, down from 15%. It still leans on one engine: Perplexity names it in 24% of its answers (36% in Edition 1), Gemini in 4%.

RE/MAX of Southern Africa’s site still returns a reproduced 403 to GPTBot while a normal browser is served, as it did in Edition 1, but it serves all three search crawlers, the ones behind each engine’s search answers. Rawson Property Group, whose site refused every visitor we tested in June, now serves every crawler. Every other tracked agency and both portals served every crawler we tested.

The engines still name a franchise’s local office in place of the national brand (“RE/MAX Living”, 26 answers), and the same two untracked agencies lead the challengers as in Edition 1: Fine & Country (38 answers) and Knight Frank (31).

Named by AI without being tracked

  • Fine & Countrynamed 38×
  • Knight Franknamed 31×
  • RE/MAX Livingnamed 26×
  • Firzt Property Groupnamed 19×
  • Tyson Propertiesnamed 15×
  • Century 21 South Africanamed 14×
  • Seeff Sandtonnamed 13×
  • Harcourts Dunnnamed 12×

The property table for Edition 2, with both crawler verdicts, is free to cite. How it was measured is in the method note.

Download the data (CSV)

Accounting, audit & tax firms

Measured 25 September 2026 (Edition 1: July 2026) · ChatGPT (gpt-4o-mini) · Claude (claude-haiku-4-5) · Gemini (gemini-2.5-flash) · Perplexity (sonar) · 36 prompts × 3 runs × 4 engines = 432 answers, 361 grounded (71 declined searches left out)

In September 2026, the South African accounting firms most often named by AI assistants were BDO (22% of 361 grounded answers), Forvis Mazars (19%) and PKF (16%), the same top three as in July, with every Big 4 firm behind them: PwC 9%, KPMG 8%, Deloitte 6%, EY 4%.

Accounting, audit & tax firms: share of AI-assistant answers naming each tracked brand, Edition 2 — October 2026, measured 25 September 2026, with the change from Edition 1 in percentage points.
#BrandShare of mentionEdition 1Change, pointsShortlist shareConviction
1BDO South Africa
22%
28%−5.6 (depends on the sample rule; not reported as movement, see the note below the table)26%79%
2Forvis Mazars South Africa
19%
20%−1.322%87%
3PKF South Africa
16%
16%0.018%75%
4RSM South Africa
14%
13%+1.118%74%
5Moore South Africa
10%
9%+1.411%83%
6PwC South Africa
9%
13%−4.613%88%
7KPMG South Africa
8%
12%−4.014%89%
8SNG Grant Thornton
8%
13%−5.0 (depends on the sample rule; not reported as movement, see the note below the table)6%78%
9Deloitte South Africa
6%
14%−7.8 (depends on the sample rule; not reported as movement, see the note below the table)11%82%
10The Beancounter
5%
8%−3.15%90%
11EY South Africa
4%
9%−4.97%87%
12Outsourced CFO
0%
1%−0.60%0%
—Xeroplatform
9%
16%−7.1 down—53%
—TaxTimplatform
3%
5%−1.3—100%

Share of mention = share of all 361 grounded answers naming the brand. Edition 1 = the restated Edition 1 share. Change = Edition 2 minus Edition 1, in percentage points; a move under 5 points is no change, and only moves of 5 or more are labelled. Shortlist share = the same measure on “best accounting firms” questions only. Conviction = of the answers naming the brand, the share presenting it as a recommended choice.

A fall of 5 points or more on Edition 2’s sample, but under 5 on Edition 1’s rule (declined searches kept). It depends on the rule, so it is not reported as movement.

Can the engines read the sites?

Accounting, audit & tax firms: whether each tracked brand’s website served the AI crawlers, by the Edition 1 method and by the search-crawler method new in Edition 2, probed 2 October 2026 (the publication-day run did not complete; see the method note).
BrandEdition 1 methodSearch crawlers (new)
BDO South Africavisiblevisible
Forvis Mazars South Africavisiblevisible
PKF South Africavisiblevisible
RSM South Africavisiblevisible
Moore South Africanot testednot tested
PwC South Africablocks GPTBot + ClaudeBot + PerplexityBot (403) (a normal browser was refused too; see the note below the table)not tested
KPMG South Africavisiblevisible
SNG Grant Thorntonvisiblevisible
Deloitte South Africavisiblevisible
The Beancounterthrottled GPTBot at measurement timethrottled OAI-SearchBot at measurement time
EY South Africavisiblevisible
Outsourced CFOvisiblevisible
Xeroserves AI bots unevenlyserves AI bots unevenly
TaxTimvisiblevisible

Edition 1 method = GPTBot, ClaudeBot and PerplexityBot, served or refused at the time of the fetch, kept so the two editions compare like for like. Search crawlers = OAI-SearchBot, Claude-SearchBot and PerplexityBot, the crawlers that feed each engine’s search answers, with robots.txt opt-outs counted and a normal-browser fetch as the control. Probed 2 October 2026 (the publication-day run did not complete; see the method note).

A normal browser was refused too, on both fetches, so this 403 is not specific to AI crawlers. The search-crawler column reports it as not tested.

“Throttled at measurement time” means a 429 that persisted with ten seconds between our requests. It describes the day we measured, not a standing policy.

The engine you ask changes the answer

Accounting, audit & tax firms: share of mention per AI engine for the most-named brands.
BrandChatGPTClaudeGeminiPerplexity
BDO South Africa0%31%25%21%
Forvis Mazars South Africa0%18%24%21%
PKF South Africa0%11%22%20%
RSM South Africa0%15%18%14%
Moore South Africa0%10%14%10%
PwC South Africa0%12%7%11%

The ChatGPT and Claude columns count only the answers where that engine searched the web: 47 and 98 of 108 here. Edition 1 could not tell which answers were searched, so these two columns are not compared across editions.

The top three held, in Edition 1’s order. BDO is named in 22% of answers against 28% in July. That is a 5.6-point fall on Edition 2’s sample, but a 3.5-point fall once the answers ChatGPT and Claude gave without searching are counted, as Edition 1 counted them. The fall depends on the rule, so it is not reported as movement. The same holds for SNG Grant Thornton and Deloitte.

The rule matters most here because of how our ChatGPT arm behaved. It searched the web on every city question (Cape Town, Johannesburg, Sandton, Pretoria, Durban, Gqeberha, the Winelands) and named local practices from its sources, none of them among the twelve tracked firms. It declined to search in 61 of its 81 answers to national questions, such as “Who are the best accounting firms in South Africa?”, and answered those from memory: BDO is named in 24 of those 61 answers, Deloitte in 19 and PwC in 17. Edition 2 counts only searched answers, so ChatGPT names no tracked accounting firm in this edition.

Xero is the one accounting move past the band on both rules: named in 9% of answers, down from 16%, with the fall on Gemini (25% to 15%) and Perplexity (19% to 6%). TaxTim, when named, is still presented as a recommended choice every time.

PwC’s site refused our normal-browser check as well as every AI crawler, so its 403s are not specific to AI crawlers this time; in Edition 1 the browser was served and PwC’s block of all three crawlers stood. The Beancounter answered GPTBot and OAI-SearchBot with 429s even with ten seconds between our requests. Xero’s site served our browser but answered the AI crawlers with server errors and redirects, which the method reports as uneven, and Moore South Africa’s site returned no response to any of our requests.

The challenger field is still crowded and local. My Accountant (41 answers), One Percent Accounting and Anlo Financial Solutions (26 each) lead the untracked names, and Nexia SAB&T, flagged in Edition 1 as a candidate for the tracked set, was named 14 times.

Named by AI without being tracked

  • My Accountantnamed 41×
  • One Percent Accountingnamed 26×
  • Anlo Financial Solutionsnamed 26×
  • Tax Consulting South Africanamed 17×
  • Thryvnamed 15×
  • Nexia SAB&Tnamed 14×
  • QuickBooksnamed 14×
  • CBO Accounting and Taxnamed 14×

The accounting table for Edition 2, with both crawler verdicts, is free to cite. How it was measured is in the method note.

Download the data (CSV)

Edition 2 · Pre-registered 3 September 2026

Edition 2 was committed before it was measured.

A recurring index is worth the delta, and a delta is only worth reading if the comparison was fixed before the result was known. So here is Edition 2 — what is measured, against what, when, and what counts as no change — published ahead of the sweep.

It opens with a correction
Edition 1 was restated on 3 September 2026. A filter now removes brand mentions that appeared only inside cited URLs, and it applies to both editions identically. Every baseline figure on this page is the restated one; the method note says what moved, and the files as first published are archived unchanged.
What is measured
The Edition 1 panel, unchanged: 3 verticals, 36 prompts each, 12 tracked brands and 2 platforms each. Copied verbatim and frozen in the repository on 26 August 2026 (commit 6c831b6). A check script fails on any changed prompt, class, region or alias; it runs before the sweep and again before publication.
Against which engines
The same four, grounded, 3 runs per prompt: ChatGPT (gpt-4o-mini), Claude (claude-haiku-4-5), Gemini (gemini-2.5-flash), Perplexity (sonar). The model ID each answer was requested from is recorded with the answer. If an engine serves a different model from Edition 1, that engine’s figures are reported as not comparable — never as movement. That check starts with this edition. Edition 1’s stored answers record only the model each engine was asked for, not the one the provider actually ran — the effective model was not captured until 3 September 2026. A provider-side swap between Edition 1 and Edition 2 therefore cannot be ruled out, only reported in the cases Edition 2 can see. Read “not comparable” as an alarm that can fire, never as an all-clear that did not. And it cannot fire on every engine. Measured against the live providers on 4 September 2026: ChatGPT and Claude return a resolved dated model (gpt-4o-mini-2024-07-18, claude-haiku-4-5-20251001), so a swap in those two is visible. Gemini and Perplexity return the alias we asked for, unchanged, so if either is re-pointed at new weights under the same name the swap is undetectable here and will not be flagged. Two of the four engines are genuinely instrumented; the other two are covered in form only, and this page will not claim otherwise.
When
The sweep was pre-registered for 15 September 2026; it ran on 25 September 2026, ten days late. Publication was pre-registered for 6 October 2026; Edition 2 published on 9 October 2026, three days late, on this page, with the drift against the restated baseline.
What counts as no change
A brand’s share of mention moving by fewer than 5 percentage points. Between Edition 1’s own three runs, the eight most-named brands per vertical already differed by up to 4.9 points in property, 4.2 in legal and 6.3 in accounting — a move inside that band is what one instrument does on one day. A rank change between brands fewer than 5 points apart is not reported as movement either.

The result publishes either way — including no change, and including movement that runs against the advice this site sells.

What this cannot show: a number that moves is not proof that anything a brand did, or anything we advised, caused it. Between editions the engines update their models and indexes, sites change, and competitors publish. The Index records what the engines said on two dates with one instrument. It does not know why, and Edition 2 will not claim to.

Edition 1 — July 2026 · The baseline

Edition 1, as restated.

Published on 5 July 2026, restated on 3 September 2026 (method note), with two crawler labels footnoted on 2 October 2026 (correction note). Every change above is measured against these tables, which are kept here as published.

Residential property & estate agencies

measured June 2026

Measured June 2026 · ChatGPT (gpt-4o-mini) · Claude (claude-haiku-4-5) · Gemini (gemini-2.5-flash) · Perplexity (sonar) · 36 prompts × 3 runs = 432 grounded answers

In June 2026, the South African estate agencies most often named by AI assistants were Pam Golding Properties (45% of answers), Seeff (35%) and RE/MAX (22%) — measured across 432 grounded answers from ChatGPT, Claude, Gemini and Perplexity.

Residential property & estate agencies: share of AI-assistant answers naming each tracked brand, Edition 1 — July 2026, measured June 2026.
#BrandShare of mentionShortlist shareConvictionAI crawler
1Pam Golding Properties
45%
58%91%visible
2Seeff Properties
35%
39%89%visible
3RE/MAX of Southern Africa
22%
20%91%blocks GPTBot (403)
4Chas Everitt International
20%
25%82%visible
5Rawson Property Group
20%
18%79%blocks GPTBot + ClaudeBot + PerplexityBot (403) (a normal browser was refused too; see the note below the table)
6Engel & Volkers Southern Africa
16%
27%80%visible
7Lew Geffen Sotheby's International Realty
13%
17%80%visible
8Jawitz Properties
13%
16%77%visible
9Just Property
10%
12%88%visible
10Harcourts South Africa
5%
6%68%visible
11Greeff Christie's International Real Estate
5%
6%81%visible
12Dogon Group Properties
5%
6%75%visible
—Property24platform
15%
—30%visible
—Private Propertyplatform
10%
—38%visible

Share of mention = share of all 432 answers naming the brand. Shortlist share = the same measure on “best estate agents” questions only. Conviction = of the answers naming the brand, the share presenting it as a recommended choice rather than a passing mention.

A normal browser was refused too, on both fetches, so this 403 is not specific to AI crawlers. The label is kept as first published so editions stay comparable. Corrected 2 October 2026: correction note.

The engine you ask changes the answer

Residential property & estate agencies: share of mention per AI engine for the most-named brands.
BrandChatGPTClaudeGeminiPerplexity
Pam Golding Properties25%55%61%38%
Seeff Properties24%37%48%30%
RE/MAX of Southern Africa13%15%34%24%
Chas Everitt International13%14%36%18%
Rawson Property Group14%12%37%17%
Engel & Volkers Southern Africa16%23%19%8%

Pam Golding Properties leads the category by a wide margin — named in 45% of all answers and presented as a recommended choice in 91% of the answers that name it. Seeff is a clear second; share drops off steeply after third place.

The assistant you ask changes the answer. Our ChatGPT arm (OpenAI's gpt-4o-mini through the API, with web search offered) named the fewest brands (Pam Golding peaks at 25% there), and that model often chose not to search: 43% of its property answers cite no source. That is the API model's behaviour, not necessarily the ChatGPT app's. Gemini names the most (61% for the same brand), and Perplexity leans on the portals — Property24 is named in 36% of its answers, within two points of Pam Golding there. No single engine is a proxy for “AI”.

RE/MAX of Southern Africa’s website returned confirmed, reproduced 403 responses to GPTBot specifically at measurement time, while a normal browser was served. Rawson Property Group’s site returned 403 to GPTBot, ClaudeBot and PerplexityBot, but to our normal-browser check as well, so that refusal is not specific to AI crawlers (corrected 2 October 2026; see the note under the table). We do not attribute either brand’s per-engine share to its crawler result: the ChatGPT arm named fewer brands across the whole panel in this sweep (see the restatement note). Every other tracked brand served all tested crawlers normally.

The engines frequently name a franchise’s local office (“RE/MAX Living”) rather than the national brand, and untracked challengers surface on their own — Fine & Country was named 40 times without being on the tracked list. Regional specialists do not automatically hold their home turf in AI answers.

Named by AI without being tracked

Every answer is open-coded, so brands outside the tracked set surface on their own — the challengers the engines already trust:

  • Fine & Countrynamed 40×
  • Knight Frank South Africanamed 27×
  • Tyson Propertiesnamed 18×
  • EXP South Africanamed 18×
  • RE/MAX Livingnamed 17×
  • Century 21 South Africanamed 16×
  • LRE Groupnamed 14×
  • Firzt Realtynamed 12×

Drift — how these rankings move between editions — begins with Edition 2. The tracked least-named brands this edition: Harcourts South Africa, Greeff Christie's International Real Estate, Dogon Group Properties.

Download the data (CSV)

Attorneys & law firms

measured July 2026

Accounting, audit & tax firms

measured July 2026

Measured July 2026 · ChatGPT (gpt-4o-mini) · Claude (claude-haiku-4-5) · Gemini (gemini-2.5-flash) · Perplexity (sonar) · 36 prompts × 3 runs = 432 grounded answers

In July 2026, the South African accounting firms most often named by AI assistants were BDO (28% of 432 grounded answers), Forvis Mazars (20%) and PKF (16%) — with every Big 4 firm behind them: Deloitte at 14%, PwC 13%, KPMG 12%, EY 9%.

Accounting, audit & tax firms: share of AI-assistant answers naming each tracked brand, Edition 1 — July 2026, measured July 2026.
#BrandShare of mentionShortlist shareConvictionAI crawler
1BDO South Africa
28%
38%81%visible
2Forvis Mazars South Africa
20%
29%79%visible
3PKF South Africa
16%
24%65%visible
4Deloitte South Africa
14%
26%75%visible
5PwC South Africa
13%
26%71%blocks GPTBot + ClaudeBot + PerplexityBot (403)
6SNG Grant Thornton
13%
12%78%visible
7RSM South Africa
13%
18%59%visible
8KPMG South Africa
12%
23%71%visible
9EY South Africa
9%
18%69%visible
10Moore South Africa
9%
12%73%not tested
11The Beancounter
8%
8%78%blocks ClaudeBot (403)
12Outsourced CFO
1%
0%50%visible
—Xeroplatform
16%
—61%serves AI bots unevenly
—TaxTimplatform
5%
—100%visible

Share of mention = share of all 432 answers naming the brand. Shortlist share = the same measure on “best accounting firms” questions only. Conviction = of the answers naming the brand, the share presenting it as a recommended choice rather than a passing mention.

The engine you ask changes the answer

Accounting, audit & tax firms: share of mention per AI engine for the most-named brands.
BrandChatGPTClaudeGeminiPerplexity
BDO South Africa18%32%31%31%
Forvis Mazars South Africa15%19%24%22%
Xeroplatform7%13%25%19%
PKF South Africa9%16%22%16%
Deloitte South Africa16%13%16%11%
PwC South Africa9%15%14%16%

The AI shortlist for accounting is mid-tier-led. BDO is named in 28% of all answers and holds the shortlist questions too (38%) — while the Big 4, the category’s biggest brands, sit mid-table and EY places ninth of the twelve tracked firms. The prompt panel asks the questions real buyers ask — small-business accounting, tax practitioners, payroll, audits for private companies — and on those questions the engines reach past the Big 4’s brand weight.

The crawler pattern from property and legal repeats at the top of the market: PwC’s website returned confirmed, reproduced 403 responses to GPTBot, ClaudeBot and PerplexityBot at measurement time while a normal browser was served — the one three-crawler block in this edition that is specific to AI crawlers — and The Beancounter blocked ClaudeBot. PwC stays on the shortlist anyway, carried by third-party sources rather than its own pages, which is the edition’s recurring lesson: what the engines can’t read, someone else describes for you.

The engine split repeats the pattern from property and legal: our ChatGPT arm (gpt-4o-mini through the API) names the fewest brands (BDO peaks at 18% there against 31–32% on Claude, Gemini and Perplexity), and 57% of its accounting answers cite no source, because that model often chose not to search. Checking one assistant still tells you little about the other three.

Software sits inside the consideration set, not beside it. Xero is named in 16% of answers — more often than KPMG — and TaxTim, when named, is presented as a recommended choice every time (20 of 20 answers), the highest conviction in the vertical. For accounting questions, the engines treat platforms as answers in their own right, not just tools.

The open capture surfaced a crowded challenger field: Sage (named 30×), Tax Consulting South Africa (25×), QuickBooks (20×) and Nexia SAB&T (20×) all register without being tracked, alongside a long tail of local practices — My Accountant, SureBooks, Anlo Financial Solutions. The SME end of the category is contested and, on this evidence, winnable.

Named by AI without being tracked

Every answer is open-coded, so brands outside the tracked set surface on their own — the challengers the engines already trust:

  • Sagenamed 30×
  • Tax Consulting South Africanamed 25×
  • Brendmo Accountantsnamed 23×
  • SureBooksnamed 22×
  • C2M Chartered Accountantsnamed 21×
  • My Accountantnamed 21×
  • Thryvnamed 20×
  • Nexia SAB&Tnamed 20×

Drift — how these rankings move between editions — begins with Edition 2. The tracked least-named brands this edition: Moore South Africa, The Beancounter, Outsourced CFO.

Download the data (CSV)

What the Index does not measure

  • Business quality. Share-of-mention measures AI visibility, not the quality of any brand’s service, people or results. A low share is not a judgement of the business.
  • Causes, beyond the confirmed ones. Absence from AI answers is multi-causal — authority, content depth, crawler configuration, name ambiguity. The Index claims a specific cause only where it is directly measured and reproduced (e.g. a 403 served to a named crawler and not to a normal browser).
  • Traditional search. Google’s ranked results are a different, separately-measurable surface. A brand can lead classic SEO and still miss the AI shortlist — and vice versa.
  • Paid placement. None of the engines measured here sold placement in these answers at measurement time; the Index measures organic answer behaviour only.
  • Permanence. Models drift. Every figure is date-stamped and holds for its measurement window, not forever — that is why the Index recurs.

Single-digit shares are noisy at N=3 and should be read as “barely registers”, not as precise rank.

Absence is multi-causal — authority, content, crawler configuration or name ambiguity. No single cause is claimed for any brand.

LegalWise and Scorpion Legal Protection are legal-expenses platforms, not law firms; they are reported in the platform bucket and excluded from shortlist share.

Xero and TaxTim are software platforms, not firms; they are reported in the platform bucket and excluded from shortlist share.

The prompt panels, in full

Every prompt behind every number on this page. Panels are fixed per vertical so editions stay comparable; each mixes shortlist, category, process, comparison and trust questions the way real buyers ask them.

Residential property & estate agencies — 36 prompts
  1. Who are the best estate agents in South Africa?
  2. What is the most reputable estate agency in South Africa?
  3. Who are the best estate agents in Cape Town?
  4. Best estate agents on the Atlantic Seaboard in Cape Town (Camps Bay, Clifton, Sea Point)?
  5. Best estate agents in Cape Town's southern suburbs (Constantia, Claremont, Newlands)?
  6. Best estate agents in Cape Town's northern suburbs (Durbanville, Bellville)?
  7. Who are the best estate agents in Johannesburg?
  8. Best estate agents in Sandton?
  9. Best estate agents in Fourways and Midrand?
  10. Best estate agents in Pretoria (Waterkloof, Brooklyn, Lynnwood)?
  11. Best estate agents in Centurion?
  12. Who are the best estate agents in Durban?
  13. Best estate agents in Umhlanga and Ballito?
  14. Best estate agents in Stellenbosch and the Cape Winelands?
  15. Best estate agents on the Garden Route (George, Knysna, Plettenberg Bay)?
  16. Best estate agents in Gqeberha (Port Elizabeth)?
  17. Which estate agency is best to sell my house in Cape Town?
  18. Best rental and letting agents in Johannesburg?
  19. Best rental and letting agents in Cape Town?
  20. Which agency is best for luxury and high-end property in South Africa?
  21. Which agency is best for selling a luxury home in Cape Town?
  22. Which estate agents are best for buying off-plan or new development property in South Africa?
  23. Best estate agent for first-time home buyers in South Africa?
  24. Who are the best commercial property brokers in South Africa?
  25. How do I choose an estate agent to sell my house in South Africa?
  26. How do I sell my house in South Africa?
  27. What commission do estate agents charge in South Africa?
  28. Should I sell my house privately without an estate agent in South Africa?
  29. What questions should I ask an estate agent before signing a mandate in South Africa?
  30. Sole mandate vs open mandate when selling a house in South Africa — which is better?
  31. Pam Golding vs Seeff — which is better to sell my house with?
  32. RE/MAX vs Rawson for selling a house in South Africa — which is better?
  33. What are the best alternatives to Pam Golding Properties in South Africa?
  34. Is Property24 or Private Property enough to sell my house, or do I still need an estate agent in South Africa?
  35. How do I check that an estate agent is registered and legitimate in South Africa (PPRA / Fidelity Fund Certificate)?
  36. How do I avoid estate agent scams when selling property in South Africa?
Attorneys & law firms — 36 prompts
  1. Who are the best law firms in South Africa?
  2. What is the most reputable law firm in South Africa?
  3. Best attorneys in Cape Town?
  4. Best law firms in Johannesburg?
  5. Best attorneys in Sandton?
  6. Best law firms in Pretoria?
  7. Best attorneys in Durban?
  8. Best attorneys in Gqeberha (Port Elizabeth)?
  9. Best conveyancing attorneys in Cape Town?
  10. Best conveyancing attorneys in Johannesburg?
  11. Best divorce and family lawyers in Cape Town?
  12. Best divorce and family lawyers in Johannesburg?
  13. Best labour lawyers in South Africa?
  14. Best commercial law firms for small businesses in South Africa?
  15. Best personal injury and Road Accident Fund attorneys in South Africa?
  16. Best criminal defence attorneys in Johannesburg?
  17. Which law firm should I use to register a trademark in South Africa?
  18. Best law firm for drafting a will and handling deceased estates in South Africa?
  19. Which attorneys are best for the property transfer when buying a house in Cape Town?
  20. Best lawyers for an unfair dismissal (CCMA) case in South Africa?
  21. Which law firm is best for immigration and visa matters in South Africa?
  22. Best law firm for debt collection for a small business in South Africa?
  23. Which attorneys are best for an antenuptial contract in South Africa?
  24. Best tax lawyers in South Africa?
  25. How do I choose an attorney in South Africa?
  26. How much does a lawyer cost per hour in South Africa?
  27. How much are conveyancing fees when buying a house in South Africa?
  28. Do I need a lawyer for a divorce in South Africa, or can I do it myself?
  29. What questions should I ask a lawyer before hiring them in South Africa?
  30. Who appoints the conveyancing attorney when a house is sold in South Africa, and can I choose my own?
  31. LegalWise vs hiring a private attorney in South Africa — which is better?
  32. Big law firm vs boutique attorney for a small business in South Africa — which is better?
  33. What are the best alternatives to LegalWise in South Africa?
  34. Scorpion Legal Protection vs LegalWise — which is better?
  35. How do I check that an attorney is admitted and in good standing with the Legal Practice Council in South Africa?
  36. How do I avoid being overcharged by an attorney in South Africa, and what can I do about a fee dispute?
Accounting, audit & tax firms — 36 prompts
  1. Who are the best accounting firms in South Africa?
  2. What are the top audit firms in South Africa?
  3. Best accounting firms in Cape Town?
  4. Best accounting firms in Johannesburg?
  5. Best accountants in Sandton?
  6. Best accounting firms in Pretoria?
  7. Best accountants in Durban?
  8. Best accounting firms in Gqeberha (Port Elizabeth)?
  9. Best accounting firms for small businesses in South Africa?
  10. Best tax practitioners in South Africa?
  11. Best tax consultants in Cape Town?
  12. Best tax consultants in Johannesburg?
  13. Best bookkeeping services for small businesses in South Africa?
  14. Best payroll providers for small businesses in South Africa?
  15. Best auditors for a small private company in South Africa?
  16. Best accounting firms in Stellenbosch and the Cape Winelands?
  17. Which accounting firm is best for a startup in South Africa?
  18. Best accountant for freelancers and sole proprietors in South Africa?
  19. Which firm is best for handling a SARS tax dispute in South Africa?
  20. Best accounting firm to prepare annual financial statements for a small company in South Africa?
  21. Which accountants are best for Xero cloud accounting in South Africa?
  22. Best firm for VAT registration and VAT returns in South Africa?
  23. Best accounting firm for company registration and CIPC compliance in South Africa?
  24. Best accountants for e-commerce businesses in South Africa?
  25. How do I choose an accountant for my small business in South Africa?
  26. How much does an accountant cost per month for a small business in South Africa?
  27. Does my company need an audit, an independent review, or a compilation in South Africa?
  28. Can I do my own tax return or should I use a tax practitioner in South Africa?
  29. What questions should I ask an accountant before hiring them in South Africa?
  30. What is the difference between a bookkeeper, an accountant, and an auditor in South Africa?
  31. TaxTim vs hiring a tax practitioner in South Africa — which is better?
  32. Big 4 vs mid-tier accounting firm for a growing South African business — which is better?
  33. What are the best alternatives to the Big 4 audit firms in South Africa?
  34. Outsourced accounting firm vs hiring an in-house accountant in South Africa — which is better?
  35. How do I check that an accountant is registered with SAICA or SAIPA in South Africa?
  36. How do I check that a tax practitioner is registered with SARS in South Africa?

How the Index is measured

Four grounded engines, a fixed prompt panel per vertical, 3 runs per prompt, and an LLM open-capture pass that reads every answer — naming each brand, judging recommended-vs-mentioned, and surfacing brands nobody tracked. Every brand’s site is also fetched as each AI crawler to test whether the engines can read it at all. The full method, including scoring rules, rounding, and known limitations, is published and versioned:

Read the full methodology →

Cite the Index

The Index is free to cite with attribution (CC BY 4.0). This page is the canonical home — editions are versioned here, never moved to new URLs.

Auto Alpha Advisory, The SA AI Visibility Index, Edition 2 — October 2026.
https://aiv.autoalphaadvisory.co.za/research/sa-ai-visibility-index

Edition 2 — October 2026 (current): the CSVs above, and the method note on the sample rule, the band and both crawler verdicts. Edition 1’s files: property, legal, accounting.

Edition 1 — July 2026, the baseline: restated 3 September 2026 (the method note states what moved, and the files as first published are archived unchanged); two crawler labels footnoted on 2 October 2026 (correction note).

Where does your brand land?

The Index tracks industries. AIV tracks your brand — all four engines, every month, with the exact pages AI cites instead of yours. Start free:

Check if AI can read your site →Get your full scorecard