By KYN AI Advisory Team — AI implementation specialists, Singapore
A competitor citation audit asks an AI assistant 50 questions your buyers would ask, then counts two things: which brands it names and which web pages it cites. The brand count shows who you are compared with. The citation count shows which pages you must be on or beat. Fifty prompts is enough to see a stable pattern for one market, and small enough to rerun every month.
Why 50 prompts and not 5?
One prompt tells you an anecdote. Answers vary between runs, so a single answer proves nothing. Fifty prompts, grouped by buyer stage, give you a distribution: a brand that appears in 30 of 50 answers is a real presence, and one that appears in 2 is noise.
Split the 50 by what the buyer is trying to do:
- 15 problem prompts. "How do I reconcile bank statements without a full-time accountant?"
- 15 category prompts. "What is a custom CRM and who builds one in Singapore?"
- 10 comparison prompts. "Custom CRM vs HubSpot for a 20-person firm."
- 10 decision prompts. "What should I ask an AI automation vendor before signing?"
Write them in the words a founder types, not the words your marketing team uses. Add the market in half of them ("in Singapore", "for a Malaysian SME") because buyers do, and the answers change with it.
What do you count in each answer?
Log every answer as one row. Five fields are enough.
| Field | What to record | Why it matters | |---|---|---| | Prompt ID | Number 1 to 50 | Lets you rerun the same set | | Brands named | Each company or product mentioned, in order | Your competitor set, as the assistant sees it | | Sources cited | Each URL or domain linked | The pages the assistant trusts | | Your position | Named, cited, both, or neither | Your share of voice | | Claim made | What it said about you, if anything | Accuracy check |
Then total three numbers per brand: the number of prompts where it is named, the number where its own site is cited, and the number where a third party writing about it is cited. The gap between the second and third is useful. A brand named often but cited only through reviews has a reputation but a weak website. A brand cited from its own site has pages the assistant can quote.
How do you run the prompts consistently?
Consistency matters more than the tool. Fix these before the first run:
- One assistant with search switched on, one account, a fresh conversation per prompt.
- The same wording every month. Change a prompt only by adding a new ID.
- The same location setting, since a Singapore query can return different sources from a Jakarta one.
- Three runs for any prompt you plan to act on, and record the majority result.
- A date on every row.
Do it by hand once to learn the shape of the answers. After that, an agent can run the set, extract brands and URLs into the sheet, and flag changes. The extraction is a narrow task with a checkable answer, which suits automation. The judgement about what to do stays with you.
How do you rank what you find?
Sort the cited domains by the number of prompts they appear in. You will usually see three layers:
- Competitor sites. Pages from the brands in your category.
- Publishers and directories. Listicles, review sites, comparison pages.
- Reference pages. Government, regulator or vendor documentation.
Each layer calls for a different action. For competitor pages, read the page that is cited and note what it has that yours lacks: a table, a price, a definition, a date. For publishers, decide whether you can be included, by a fair submission or a data point they would want. For reference pages, link to them from your own content, because that is what the assistants are already citing.
Then score each gap by two numbers: how many prompts it affects, and how close the buyer is to a decision. A decision-stage prompt where you are absent outranks a problem-stage prompt where you are absent.
What are the traps in a citation audit?
- Treating one run as a ranking. Answers are sampled. Report counts across runs, not a position.
- Counting brand mentions in your own prompts. If your prompt names a competitor, the answer will too. Keep prompts neutral.
- Ignoring inaccuracy. If the assistant states a wrong price or capability for you, that outranks a missing mention. Fix the source page.
- Chasing every gap. Pick the five that cover the most prompts.
- Auditing once. A single audit is a photo. Monthly runs show whether your changes moved anything.
For the wider method on getting pages cited, see how to get a business website cited by ChatGPT and Google AI Overviews and appearing in AI search answers.
What does it cost to run?
Time and a small model bill. By hand, 50 prompts with logging takes a working day. With an agent, the run and the extraction are automated and a person reviews the sheet. KYN's own AI search check costs about S$0.05 per check, and we run 22 checks. That figure depends on the model and the length of the answers.
What does a worked example look like?
The numbers below are illustrative. They show the shape of a result, not a benchmark.
A 15-person firm that sells payroll software in Malaysia runs its 50 prompts and finds the following across 50 answers:
| Brand | Named in | Own site cited in | Third party cited in | |---|---|---|---| | Competitor A | 31 | 9 | 14 | | Competitor B | 24 | 15 | 6 | | Competitor C | 12 | 2 | 10 | | Your firm | 3 | 0 | 2 |
Read the table in three passes. First, Competitor A is named most, yet its own site is cited in only 9 answers, so its reputation outruns its pages. Second, Competitor B has the strongest site: 15 citations of its own pages out of 24 mentions. Its pages are the ones to read closely. Third, your firm is named in 3 answers and never cited from its own site, which means the assistant has no page of yours it is willing to quote.
The next step is concrete. Open the 15 pages cited for Competitor B, list what they share (a pricing table, a statutory-rate section with a regulator link, a dated "checked" line), and check which of those your own pages lack.
Which prompts should you rerun first?
Not all 50 are equal. After the first audit, mark each prompt with two scores: how many competitors appear, and whether you appear. Rerun these first each month:
- Decision-stage prompts where you are absent.
- Prompts where an answer states something wrong about you.
- Prompts where your name appeared last month and vanished this month.
The third group is a warning sign. Something changed on a cited page, or a competitor published a page that now outranks yours as a source. Compare the cited URLs from the two months before you change anything.
Where KYN fits
KYN built this audit for its own marketing: 50 buyer prompts, a counter for brands and cited sites, and a ranked findings list that feeds the article plan. Agents run in their own restricted accounts and every action is logged with the actor. Anything that publishes, sends or deletes waits for a named person. The system runs in a cloud account in your name.
The first 14 days are a review of your stack and interviews with your team. A full build ships in under 90 days. See marketing systems and the free AI search check.
Frequently asked questions
How often should I rerun the 50 prompts?
Monthly is enough for most SMEs. Rerun sooner after you publish a batch of pages or change a key page, so you can see whether the counts moved.
Do I need a paid tool to do this?
No. A spreadsheet and an assistant with search are enough to run it by hand. A paid tool or a custom agent saves time once you want the monthly rerun to happen without someone remembering to start it.
What counts as a citation?
A citation is a link to a page the assistant used as a source. A mention is a brand name in the answer text. Log them separately, because a brand can be mentioned without being linked, and the link is the part that sends a buyer to a site.
Checked 30 October 2026.
Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.
