A competitor comparison in AI answers has two halves: the comparison (per question and per app, who is named, how often, where and in what tone) and the “why” (which pages the AI leaned on for a rival, and what they have that yours lack). Below is the method for both, the real hours by hand, and which tool type covers which half.
Key points
- One answer proves nothing. In SparkToro’s January 2026 study, there was less than a 1 in 100 chance, in ChatGPT and Google’s AI, of the same brand list in any two answers; SparkToro advises asking “at least 60-100X” and averaging.
- The why is on the cited pages. Open the pages an answer cited next to the rival and check them against your matching page.
- Other sites decide a lot. A 2025 University of Toronto study found AI search took 81.9% of its sources from earned, third-party media in US car queries, against 45.1% for Google.
- Tool types split the work. Tracking-only tools cover the comparison and leave every fix to you; Far & Wide’s Audit and Fix plans measure and make the changes.
- By hand costs €0 and 22 hours a month (23 hours 15 minutes in month one) for 20 questions in 2 apps, before any change.
What’s the best tool to compare my brand against competitors in AI answers?
The best tool runs the whole loop: it asks a fixed list of buyer questions many times in the AI apps your buyers use, reads every answer for you and each rival, shows the sites each answer cited, and turns what it finds into changes on your site. Tracking-only tools stop after the reading. Far & Wide’s Audit and Fix plans cover all of it: up to 20 competitors you name, share of voice, the tone of every mention, every answer word for word, a weekly audit of your own pages, and agents that prepare each fix, publish it after your OK, check it is live and re-measure its questions.
How does AI competitor tracking work?
AI competitor tracking is the repeated measurement of which brands AI apps name for your buyer questions, and how often, compared with yours. It works by asking the same unbranded buyer questions in each AI app on a fixed rhythm, many times each, and recording for every answer who is named, where, in what tone and with which sources. The counts are added up per question, per app and per competitor, and compared over time.
- A fixed question list of real buyer questions that do not name you, such as “best accounting software for freelancers in the Netherlands”. Branded questions are kept apart, since they return your name almost every time. See how to find the prompts your customers use.
- Fixed conditions: the same apps, your buyers’ country and language, a fresh chat each time, a fixed day.
- Many asks per question. In SparkToro’s study, 600 volunteers ran 12 prompts a combined 2,961 times, and two lists came out in the same order only about 1 in 1,000 runs.
- The same fields from every answer: named or not, place, rivals named instead, tone of each mention, cited URLs.
- Totals, not ranks. SparkToro calls visibility % across many prompts “run multiple times” a reasonable metric, and says a “ranking position in AI” is “full of baloney”. For the formulas, see how to measure AI share of voice.
The cited URLs are the input for the “why” half, so a tracker that does not log them can tell you that you lost, but not why.
How do I compare my brand against competitors, step by step?
You compare your brand against competitors by running the same 20 questions in at least two AI apps, three times each per round, and logging one row per answer. The effort pays where buyers ask AI directly: a 2025 Bain report found 42% of LLM users ask them for shopping recommendations.
1. Write 20 unbranded buyer questions
Cover the buying stages: “best [category] for [audience]”, “[Rival] alternatives”, “[Rival] or [your brand]?”, “cheapest [category] with [feature]”. Freeze the wording.
2. Ask each question 3 times per app, in fresh chats
Use the ChatGPT and Gemini apps at minimum, and note whether the app searched the web: an answer with sources can be traced to pages.
3. Log one row per answer
Give every answer its own row, with these columns:
| Column | What to record |
|---|---|
| Named | Yes or no |
| Place | First, second, later |
| Rivals named | Every brand listed |
| Tone | Positive, neutral or negative, for you and each rival |
| Cited URLs | Every link, with the rival it supported |
| Full answer | Word for word |
4. Sort presence problems from preference problems
A presence problem means you are rarely named; you need pages and mentions AI can find. A preference problem means a rival comes first or gets warmer words; you need better reasons on the pages AI already reads. Mark the 10 questions where a rival wins most clearly.
How do I find out why AI picks my rivals over me?
You find out why AI picks a rival by opening the pages the answer cited when it named that rival, and checking each one against your own page for the same question. The reason is usually concrete: a price written out, a comparison table, a review site you are missing from, or text the AI can read on their page and cannot read on yours.
General reasons are in why competitors appear in ChatGPT and you don’t and how AI chooses brands to recommend; the method below goes rival by rival.
5. Open the two most-cited pages behind each winning rival
For each of the 10 questions, take the two URLs cited most often next to the rival: usually its own page and a third-party page such as a review-site listing. Allow about 15 minutes per page to read it, run the checks and compare it with yours.
6. Run the six checks on each page
Run the same six checks on your own matching page, then compare the two:
| What the rival page has | How to check | What to change on yours |
|---|---|---|
| A direct answer in the first lines | Do the first 2 sentences answer the question? | Answer first, details after |
| A comparison table with prices | A table, with euro or dollar amounts | Publish prices and a table, not “contact us” |
| A page on a third-party site | Is the cited page on another domain, and are you on it? | Get onto the same review sites and publications |
| Text visible without JavaScript | Is the answer in the raw page source? | Move text out of tabs, sliders, popups and images |
| Crawler access | robots.txt and CDN rules for OAI-SearchBot | Allow OAI-SearchBot; see the llms.txt guide |
| Structured data matching the page | Compare JSON-LD with the visible text | Fix mismatches; see schema markup for AEO |
Crawler access is a hard stop. OpenAI states that sites opted out of OAI-SearchBot “will not be shown in ChatGPT search answers”, and a robots.txt change takes about 24 hours to apply. For AI Overviews and AI Mode, Google requires a page “indexed and eligible to be shown in Google Search with a snippet”, and recommends content “in textual form” and structured data that “matches the visible text on the page”.
Third-party pages often decide it. The Toronto study found a “systematic and overwhelming bias towards Earned media” in AI search; your own page often cannot win alone against a review site.
A worked example. Asked “best accounting software for freelancers in the Netherlands”, an app names [Rival] first and cites two pages: [Rival]’s pricing page, which opens with its monthly price over a three-plan table, and a review-site roundup that lists [Rival] but not you. Your pricing page says “pricing on request” and keeps features in tabs. The diagnosis is three changes: write out your prices, add a comparison table in plain HTML, and get onto that review site.
7. Turn each gap into a change, and re-measure
List each gap with the questions it should move, make the change, and ask those questions again in the next rounds. Allow about 1 to 2 hours for a rewritten page and 3 to 4 hours for a new page. Code changes such as robots.txt, structured data or text moved out of sliders need someone who can edit the site, or a platform that ships the code for you.
What are the best tools to check my competitors’ AI visibility?
The best tools to check competitors’ AI visibility fall into three types: the AI apps and a spreadsheet by hand, a tracking-only tool, and a platform that measures and also makes the changes. Tracking-only tools cover the comparison half; Far & Wide, on Audit and Fix, is the platform type, covering the comparison, the cited sources and the fixes.
| By hand (AI apps + spreadsheet) | Tracking-only tool | Far & Wide Audit and Fix (measures and makes the changes) | |
|---|---|---|---|
| What it compares | You vs the rivals you log; 20 questions, 2 apps | You vs the rivals you set, per question and app | You vs up to 20 competitors you name and other brands AI names instead; 50 questions, 4 apps you pick of 6; share of voice; tone; every answer word for word |
| Asks per question per app a month | 12 (3 a week × 4 weeks) | Varies by tool | 60 by arithmetic (3 personas × 5 asks × 4 rounds); fewer on Google AI Overviews, which gives one answer however often it is asked |
| Explains why the rival won? | Only if you open and compare the cited pages (5 h a month) | Lists cited sources; comparing pages is your job | Cited sites sorted into 9 kinds, competitor sites and your own pages included; weekly audit of your pages: every problem, why it matters to AI, what to change |
| Makes the changes? | No, you make every change | No, every fix is your job | Yes: agents prepare code, page fixes, new pages and outreach; after your OK it is published (GitHub pull request, or a draft in WordPress, Wix (blog posts), Webflow or HubSpot; ready-to-paste code elsewhere), checked live, kept with an undo and re-measured; outreach sent after your OK, placement never guaranteed |
| Your hours a month | 22 h (23 h 15 min in month one) before any change; 28 to 32 h with 3 page rewrites and 1 new page | Reading reports, the why step and every change | Approving finished changes, a few minutes a week |
| Cash cost | €0 | A monthly subscription | Audit €99/month excl. VAT (7-day free trial, no card); Fix €299/month excl. VAT |
Choose by hand if you track under 20 questions, have someone who can edit the site, and can spare 22 hours a month for the measuring alone and 28 to 32 hours once you make the changes, at €0 cash. Choose a tracking-only tool if you only need the comparison and have people for every fix. Choose a platform that makes the changes if you want the gaps closed each month, not just found. More on the category: AEO tracking software.
What do Far & Wide’s monthly plans include?
Far & Wide has three monthly plans, Monitor, Audit and Fix, priced excl. VAT. Every plan includes 50 buyer questions, the 4 AI apps you pick of 6 (ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Overviews, Google AI Mode), every answer word for word, up to 20 competitors and share of voice, the tone of every mention and trends over time.
| Plan | Price excl. VAT | How often | Personas × asks | AI answers a month | Site audit | Fixes a month | Trial |
|---|---|---|---|---|---|---|---|
| Monitor | €15/month | Every 7 days | 1 × 3 | Up to 2,400 | None included; one-off €49 | None | No |
| Monitor every 3 days | €30/month | Every 3 days | 1 × 3 | Up to 6,000 | None included; one-off €49 | None | No |
| Audit | €99/month | Every 7 days | 3 × 5 | Up to 12,000 | Weekly, 15 main pages | 1 try-out in each area | 7 days free, no card |
| Fix | €299/month | Every 7 days | 3 × 5 | Up to 12,000 | Weekly, up to 100 pages | A large number of technical, 5 web design, 10 content articles, 20 authority | No |
AI answers are “up to” because Google AI Overviews gives one answer however often it is asked. Monitor covers AI answers only: no site audit included (a full audit of 15 pages can be bought for €49 excl. VAT), no fixes, no trial. Fix includes everything in Audit, on up to 100 pages.
The fixes come from four groups of agents: Technical (crawler access, structured data, titles and descriptions, your sitemap and llms.txt, as exact code), Web Design (content hidden in accordions, sliders and images), Content (corrects wrong or weak content and writes what is missing: answers to buyer questions where AI names nobody, comparison pages, articles) and Authority (outreach ready to send). After your OK each fix is published (a GitHub pull request, or a draft in WordPress, Wix (blog posts), Webflow or HubSpot; ready-to-paste code elsewhere), checked live and kept with an undo; outreach is sent after your OK, and placement is never guaranteed. See the pricing page.
How long does a competitor comparison take by hand?
A competitor comparison by hand takes 22 hours a month, and 23 hours 15 minutes in month one, for 20 questions in 2 apps, before you change anything. The cash cost is €0. The hours add up as follows:
- Set-up, once: 1 h 15 min. 20 questions (45 min) and the log sheet (30 min).
- Comparison rounds: 16 h a month. 20 questions × 2 apps × 3 asks = 120 answers a week, at about 2 minutes each to open a fresh chat, ask, read and log = 4 h a week, × 4 weeks.
- The why step: 5 h a month. 10 questions × 2 cited rival pages × 15 minutes.
- Monthly review: 1 h.
- Changes, on top: 1 to 2 h per rewritten page, 3 to 4 h per new page. Three rewrites and one new comparison page add 6 to 10 h, for 28 to 32 h a month. Code changes and outreach to review sites come on top of that, and need someone who can edit the site and someone to pitch the sites.
Each extra app adds 8 hours a month. The routine gives 12 answers per question per app a month against SparkToro’s 60 to 100, cannot change code without someone who can edit the site, and does not keep up a monthly output of pages and outreach. In hours, by hand is the most expensive option here; only its cash cost is lower.
Common mistakes
- Judging from one screenshot: one answer is noise.
- Tracking rank position instead of how often you are named, and named first.
- Fixing your homepage when the cited page is a review site.
- Changing pages without re-asking the same questions.
Quick checklist
- 20 unbranded buyer questions, wording frozen
- At least 2 AI apps, 3 asks per question per round, fresh chats
- Named, place, rivals, tone, cited URLs and full answer logged
- Two cited pages per winning rival run through the six checks
- Each gap turned into a change, and its questions re-asked
FAQ
A tracker asks a fixed list of unbranded buyer questions in each AI app, from your buyers’ country and language, on a fixed rhythm, several times each. For every answer it records whether you are named, in what place, which rivals appear instead, the tone of each mention and the cited URLs, then totals them per question, app and competitor over time.
Measure your rivals, and get the changes made
Comparing 20 questions by hand takes 22 hours a month and leaves every change to you. On Audit and Fix, Far & Wide asks your 50 buyer questions every 7 days in the 4 AI apps you pick, reads every answer for you and up to 20 competitors, shows the cited sites and audits your pages weekly. Its agents prepare the fixes (on Audit, one try-out a month in each area; on Fix, a large number). After your OK each fix is published, checked live and kept with an undo, and its questions are measured again at the next audit.
Try Audit free for 7 days, no card needed: add a card before the trial ends to keep it at €99/month + VAT, or let it end and pay nothing. Monitor, from €15/month + VAT, tracks the AI answers only; Fix, at €299/month + VAT, starts straight away.
Start 7-day free trial