Auto Alpha AdvisoryAsk an owner who their competitors are and the answer comes back in seconds: the three firms they lost the last few deals to. That list is real, but it's yours. The market keeps its own, and the gap between the two is usually the first finding of a competitor website analysis.
I'm a CA(SA), and I read competitors for a living, in the Competitive Position door of The Full Read. This is that method cut down to an afternoon, with an honest note on where it breaks. As for whether AI can do most of it: it can do most of the typing. Claude or ChatGPT will summarise and tabulate a handful of competitor sites in minutes. It can't tell you who the market names, how those sites perform or whether their claims are true, and that's where the findings are.
Start with the market's list, not yours
Write down your three. Then take your buyers' top three questions, the ones they'd type before they ever call you (who's a good accountant for a small business in Cape Town? is the shape), and put each one to four AI assistants: ChatGPT, Claude, Gemini and Perplexity. Ask each question several times, in a fresh chat every time, and write down every business that gets named.
The repeats matter because the answers move. Aleyda Solis's BrightonSEO deck of 30 April 2026 puts it plainly: "Look for patterns over time, not single-run results, because AI outputs vary by session and platform." When David Kaufman put one software question to ChatGPT 100 times for Siteline in April 2025, even the most-named product appeared in only 94% of the answers. The SA AI Visibility Index asks each question three times on each of four models for the same reason. Three questions, four assistants, three runs: 36 answers. That's a small sample, but it is a sample.
Now set that tally beside your three. It's common for a firm you never listed to keep turning up while one of yours barely appears. That doesn't make your list wrong. Yours is who you meet at the pitch; the tally is who a buyer hears about before there is a pitch. The difference tells you where to look first.
Read every competitor on the same day, with the same questions
Take the three or four most-named firms from both lists and read each site, and your own, across the same five areas:
- What the site claims and offers: what they say they do, for whom, and what you can buy.
- What proof it shows: named clients, case studies, credentials, reviews, real people with qualifications.
- How fast it loads: open each homepage on the same phone, on the same connection, in the same hour, and note how long it takes before you can read the headline.
- Whether the content is in the page without JavaScript: open View Page Source and search it for the headline and the service list. If they aren't in the source, the browser is building them.
- Who else on the web mentions them: search the name in quotes, exclude their own domain, and note which directories, press, associations and review sites come back.
For each area, ask the same three things: what are they doing, what's working for them, and where's the gap? Be strict about the middle one. Working should be something you can point to, like the firm turning up in your tally or winning the deals you lose. A nice-looking site doesn't count.
Area four is the one owners skip. Google's JavaScript documentation describes a headless Chromium that "renders the page and executes the JavaScript". AI crawlers are another matter. Vercel and MERJ's study, The rise of the AI crawler, published on 17 December 2024, covered nearly 1.3 billion fetches in a month from four AI crawlers on Vercel's network. Its conclusion: "The results consistently show that none of the major AI crawlers currently render JavaScript." If a rival's services are in their source and yours aren't, that's a fix, not a strategy problem.
The sheet for each firm is a table, with the five areas as rows and the three questions as columns. There's a filled-in one below, written out row by row. Read every firm on the same day. Sites change, and a rival read this afternoon against your own site from last month isn't a comparison.
Run the free scan on your own site, not theirs
For the technical side of your own site, run the free scan. Its nine domains run from security to AI readiness (GEO), and it goes well past what a phone and View Page Source can tell you. Don't point it at a competitor. Its terms require you to be entitled to have the site assessed, for example because you own or run it, or its owner has asked you to.
For rivals, stick to what any visitor can see. That makes your own technical read deeper than theirs, so keep the comparison level: the phone test and the source check go on every sheet, yours included, and the scan results go on your own fix list.
A worked example (hypothetical)
This one is invented, to show a sheet filled in. A Cape Town accounting practice writes down the three firms it loses work to. After 36 answers, two of the three show up regularly, one hardly at all, and a fourth firm the partners have never discussed is named more than anyone. The practice itself appears a handful of times. Read the same afternoon, the fourth firm's sheet looks like this:
- Claims and offer: three services, each on its own page, each saying who it's for. It works because the wording matches the buyer questions. The gap: the practice lists twelve services on one page.
- Proof: partners named with their qualifications, and two sector case studies. It works because a buyer can see who they'd be dealing with. The gap: the practice names nobody.
- Speed: the headline is readable almost at once on mobile data, so there's nothing to wait through. The gap: the practice's homepage loads a large video first.
- Content without JavaScript: the headline and services are in the source, so crawlers that don't run JavaScript can read them. The gap: the practice's service list is built in the browser.
- Mentioned elsewhere: a professional body's directory and two local business articles, which describe the firm the way it describes itself. The gap: the practice appears on its own site and one directory.
The proof and source gaps are fixes the practice controls, and neither would have turned up from reading only the three firms the partners named.
Where AI helps, and where it makes things up
Give Claude or ChatGPT the competitor pages, pasted in or as URLs, and ask for a table of what each one claims, offers and shows as proof. It will fill that in within minutes. Use it for that.
Then remember what it's reading. It reads what competitors say, not how they perform: it can't time a page on a phone, and a promise of fast turnaround goes into its table as though it were a fact. It can also state capabilities a site doesn't have, a service the firm never mentions written in with the same confidence as the ones it does. Check every line against the competitor's own page before you act on it. If you can't find the sentence on their site, delete the cell.
A single AI answer is one draw, not a ranking. Here's one instrument on one day. In the October 2026 edition of the SA AI Visibility Index, measured on 25 September, six South African law firms shared the top of the answers, from Adams & Adams at 16% to Cliffe Dekker Hofmeyr at 10%. None had moved 5 points since July, and the Index's legal section is careful about what that means: "No firm moved 5 points, so under the pre-registered rule the order among them is no change." That's with 378 answers behind it. Calling one of those six the leader on a single check would be reading noise. Your 36 answers deserve the same caution: use them to see who's in the conversation, not to rank anyone.
Don't add the areas into one score
Once the sheets are done, the temptation is to mark each area out of ten and add them up. Don't. A rival well ahead on proof and behind on speed can total the same as you, and the sum then says you're level when you aren't. The gap you came for sits in one row. Keep the rows apart and rank the gaps, not the firms.
What the afternoon can't give you
An afternoon gets you the market's list and a sheet per firm. The Competitive Position door of The Full Read runs the same method in full: three named competitors read on the same instruments on the same day, their public claims checked against sources, who four AI models name in your category across a fixed panel of buyer questions, and a plan ordered by the gap to the leader. For three competitors it takes about a week. It's a read, not an audit, with the evidence beside each finding so you can check it yourself.
Do the afternoon first. If the gap between your list and the market's is bigger than you expected, that's worth a conversation. The first call is 15 minutes.