AI Search Audit Deliverables: A Report Template, the Metrics and a One-Page Summary
AI search audit deliverables: an eight-section report, four core metrics and a one-page summary template you can copy. With published audit prices.
Muhammad Ramadiansyah · Co-Founder & CTO Intura10 min read

A complete AI search audit delivers an eight-section report, four core metrics and a one-page summary for management. Below you will find the report template, how to calculate each metric, an executive summary you can copy, and the questions to put to any audit provider.
This is for agency leaders packaging a GEO audit, and for marketing directors who want to know what is reasonable to ask a vendor for. The template here is one example, built from the audit reports we produce. It is not an industry standard; there is no official standard for AI search audits yet. For contrast, NEORIX, a GEO provider in Indonesia, lists its audit deliverables on its service page as a one-page executive report, a 20–30 page detailed report, a gap analysis and a 30-60-90 day roadmap.
Provider prices and package contents were checked on the providers' own pages on 21 November 2026. The template uses no client or brand data.
What are the deliverables of an AI search audit?
A complete AI search audit delivers eight things: an executive summary, the prompt list, brand mention figures, a map of the sources AI cites, an information accuracy check, a topic map, a prioritised action plan and a website readiness audit. Without the action plan, you get a snapshot, not a work plan.
The two parts most often missing from audit reports are the last two. Mention figures without an action plan tell the team where it stands but not what to work on next week. A website audit without AI answer data has the team fixing pages that AI may never read for the questions that matter.
What does an AI search audit report template look like?
The template has eight sections, and each answers one question from management: how visible are we, what was tested, who does AI recommend, who does AI learn from, what is wrong, where can we win, what do we do first, and can AI read our website.
| Section | Minimum content | Question it answers |
|---|---|---|
| 1. Executive summary | Visibility score, the four core metrics, the weakest audience segment, three priorities | How visible is our brand in AI? |
| 2. Prompt list | Every question tested, with audience, funnel stage, intent, AI engine and test date | What was tested, and can it be re-run? |
| 3. Brand mentions | Number of answers that mention the brand, mention order, competitors mentioned alongside | Who does AI recommend in our category? |
| 4. Citation map | Sources AI cites, grouped as own site, competitors, media and other, per AI engine | Who does AI learn about our category from? |
| 5. Accuracy and completeness | Every AI claim about the brand checked against official facts: correct, outdated or wrong. Plus key facts that were not mentioned | What is wrong or missing? |
| 6. Topic map | Status of each topic: owned, winnable, crowded or empty. Plus the queries AI ran while looking for an answer | Which topics can we win most cheaply? |
| 7. Action plan | Pages to create or fix, their format, channel, the reason for the gap and priority | What do we do first? |
| 8. Website audit | Per-page scores for content structure, SEO readiness and technical health. Issues ranked by impact | Can AI read and cite our website? |
Ask for one appendix too: the raw answers. That is the full text of every AI answer, with date, engine and cited links. Without it you cannot check the figures in sections 3 to 5, and you cannot compare a re-test fairly.
Which metrics should an AI search audit report include?
At least four: brand mentions, citations, information completeness and information accuracy. The first two measure whether you show up. The last two measure whether AI tells your story correctly, and both need a list of official facts from you: prices, licences, addresses and product features.
| Metric | How to calculate it | Why it matters |
|---|---|---|
| Brand mentions | Answers that mention the brand at least once, divided by all answers | If the brand is not mentioned, nothing else in the answer works for you |
| Citations | Every link to your own domain across all answers; one answer citing three pages counts as three | Citations send visitors and mark you as a source |
| Information completeness | List the facts that should appear, then count how many do; a partly stated fact scores half | Every missing fact is a page you can publish |
| Information accuracy | Check every claim against official facts: correct 100, outdated 50, wrong 0, then average | Wrong information keeps spreading until its source is fixed |
Put the calculation rules in the report, not only the figures. Two vendors can both report "12% citations" and mean different things: one counts answers with a citation, the other counts links. Choosing metrics for monthly reporting is covered in how to measure GEO performance (in Bahasa Indonesia).
What does a one-page executive summary look like?
One page is enough: four figures, three findings, three 30-day priorities and the limits of the data. The four figures are brand mentions, citations to your own site, information completeness and information accuracy. If the data limits section is missing, ask for the sample size.
Copy this template and fill in the square brackets:
EXECUTIVE SUMMARY — AI SEARCH AUDIT
Brand: [brand name] Test date: [date]
Scope: [number] prompts × [number] AI engines × [number] repeats = [number] answers
Country and language: [country], [language]
FOUR FIGURES
Brand mentions .............. [number]% ([number] of [number] answers)
Citations to own site ....... [number] links in [number] answers
Information completeness .... [number]/100
Information accuracy ........ [number]/100
THREE FINDINGS
1. [finding] — evidence: [prompt, engine, date]
2. [finding] — evidence: [prompt, engine, date]
3. [finding] — evidence: [prompt, engine, date]
THREE 30-DAY PRIORITIES
1. [action] — owner: [name] — due: [date]
2. [action] — owner: [name] — due: [date]
3. [action] — owner: [name] — due: [date]
DATA LIMITS
- Answers in the smallest group: [number]; 95% range (Wilson): [number]%–[number]%
- Engines, countries and languages not tested: [list]
- AI answers change over time; this is a snapshot as of [date]
- Re-test scheduled: [date]
How many prompts and AI engines are enough for an audit?
As a practical rule, at least 30 answers for every group you want to draw a conclusion about; below that, the figure is only an indication. The number of answers is prompts times AI engines times repeats. Ten prompts on two engines give only 20 answers, while 25 prompts on five engines give 125.
The table shows why. If your brand is never mentioned, the true mention rate is not necessarily zero. The fewer the answers, the wider the range.
| Answers | Test result | True rate could be |
|---|---|---|
| 4 | 0 mentions | 0% to 49% |
| 20 | 0 mentions | 0% to 16% |
| 30 | 0 mentions | 0% to 11% |
| 125 | 0 mentions | 0% to 3% |
Method: 95% Wilson confidence interval for 0 mentions in n answers.
A "group" is any slice you want to report on its own: one engine, one audience segment or one funnel stage. If a report splits 20 answers across five engines, each engine has four answers, and the first row of the table applies. For routine monitoring, see how many prompts to track.
How do you audit your brand in ChatGPT yourself?
You can run a small version of the audit yourself with seven steps and one spreadsheet: collect real customer questions, group them by stage, run them in fresh sessions on each AI engine, record every answer, check claims against official facts, then calculate the four metrics and write the summary.
- Write your list of official facts. Prices, licences, addresses, contacts and key product features, each with the date it applies from. This is the answer key for the completeness and accuracy metrics.
- Collect real customer questions. Take them from sales, customer service chats, reviews and forums. See how to find the questions customers ask AI.
- Group them by stage. Discovery ("best X for ..."), consideration ("X or Y?") and decision ("X price", "is X safe?").
- Run them in fresh sessions. Use a new chat with no history and, where possible, no account memory, on every AI engine you test. Repeat each prompt at least twice, because answers can differ between attempts.
- Record every answer as it is. Copy the full text and every cited link into the spreadsheet. Do not summarise yet.
- Check claims against official facts. Mark each claim about your brand as correct, outdated or wrong, and note key facts that were left out.
- Calculate the four metrics and write the summary. Use the calculation rules in the metrics table above, then fill in the one-page summary template.
A spreadsheet header you can use:
date,ai_engine,prompt,audience,funnel_stage,repeat,brand_mentioned,mention_order,competitors_mentioned,cited_urls,own_domain_citations,claim,claim_status,facts_not_mentioned,notes
If the results show your brand is rarely mentioned, the next step is in how to get your brand mentioned in ChatGPT. If the problem is wrong information, read how to fix wrong brand information in ChatGPT and Gemini.
How is an AI search audit different from an SEO audit?
An SEO audit checks whether your pages can be found and ranked by Google. An AI search audit checks whether your brand is mentioned and cited inside AI answers. You need both, because AI draws its sources from the same web, and a high website score does not automatically mean AI cites you.
| SEO audit | AI search audit | |
|---|---|---|
| What is tested | Pages and keywords | Customer questions and AI answers |
| Main output | Rankings, impressions and clicks | Brand mentions, citations, completeness and accuracy |
| Data source | Site crawler, Google Search Console, keyword tools | Answers collected from each AI engine |
| Competitors | Sites ranking above you | Brands AI recommends and the domains it cites |
| Stability | Fairly stable day to day | Can differ between attempts, so it needs repeats |
How much does an AI search audit cost, and how long does it take?
Published AI search audit prices in Indonesia range from IDR 1.5 million (about US$82) for a small consultation to IDR 45 million (about US$2,460) for a project-scale audit. NEORIX says its audit takes 1–2 weeks. Run it yourself and the cost is your time plus a monitoring tool, from about US$29 a month.
| Provider | Audit format | Price |
|---|---|---|
| NEORIX | Basic consultation: 2 AI engines, 5–10 recommendations, 60-minute session | IDR 1,500,000 (≈ US$82) |
| NEORIX | Advanced consultation: 5 AI engines, 90-day roadmap, 2 sessions | IDR 3,500,000 (≈ US$192) |
| Hashmeta Indonesia | GEO audit and AI visibility assessment, one-off project | IDR 15,000,000–45,000,000 (≈ US$821–2,464) |
Checked on each provider's page on 21 November 2026. Hashmeta describes its prices as indicative. US dollar figures use the JISDOR rate of 17 November 2026, IDR 18,264 per US dollar.
The monitoring tool figure is the cheapest monthly-billed plan in the IndustryLens price index (Otterly.AI Lite, US$29 a month). For broader search service costs in Indonesia, see monthly SEO service costs in Indonesia.
What should you ask an audit provider?
Ask six things before you pay: the number of prompts, AI engines and repeats; where the prompts come from; whether you get the raw answers; how accuracy is scored; what the action plan looks like; and when and how the re-test runs. The answers show whether you will get a work plan or just a set of numbers.
- How many prompts, AI engines and repeats? Multiply the three. If the result is under 30 for a group you care about, treat its figure as an indication.
- Where do the prompts come from? Real customer questions are more useful than keywords rewritten as questions.
- Do you get the raw answers? Without them you cannot check the figures or repeat the test the same way.
- How is accuracy scored? Ask for the list of facts used as the reference and who judges each claim.
- What does the action plan look like? Ask for an example: which pages, which format, which channel, and why that comes first.
- When and how does the re-test run? It must use the same prompts, engines and country so the figures can be compared.
What does an AI search audit not show?
An audit does not show sales, and it does not guarantee that tomorrow's AI answers match today's. It is a snapshot on one date, only for the engines, countries and languages tested. Re-test after you publish fixes, so you have before and after figures.
- Not a sales figure. Being mentioned by AI is not the same as being bought. Connect it to traffic and lead data separately.
- Not a forecast. AI models and their search sources change without notice, so answers can shift at any time.
- Only the scope tested. Results for ChatGPT in Bahasa Indonesia do not hold for Perplexity in English.
- Not what every person sees. Users with account memory, a different location or chat history can get different answers from a clean test session.
If you want a report with the eight sections above for your brand or your clients, see our AI visibility audit. To see where you stand first, start with a free AI visibility audit.
Sources
Frequently asked questions
Does an AI search audit need access to your website or an API integration?
Not necessarily. Prompts can be run directly on each AI engine, with nothing installed on your website. What an audit provider needs is your list of official facts, a list of competitors and your website URL for the website audit. Google Search Console access helps compare against search data, but it is optional.
How is an audit different from monitoring tools like Otterly.AI or Profound?
Monitoring tools run the same prompts on a schedule and show trends in mentions and citations. An audit is one deeper pass: an accuracy check of every claim, a citation map, a topic map and an action plan. Many teams use an audit first to choose the prompts that matter, then a tool to monitor them.
How long does an AI search audit take?
NEORIX says its audit takes 1–2 weeks. The time depends mostly on how many answers have to be read and checked one by one: 25 prompts on five engines with two repeats is 250 answers. A small audit you run yourself can be faster, but its sample is smaller too.
How often should an audit be repeated?
Re-test about 30 days after your fixes go live, with the same prompts, engines and country, so you have before and after figures. After that, repeat each quarter or whenever something big changes, such as new prices, a new product or a rebrand. Between audits, a monitoring tool keeps the data running.
Is a free AI visibility audit enough?
It is enough to see whether your brand shows up at all. For budget decisions, check the sample size first: below 30 answers per group, the figure is only an indication. Free audits also rarely include an accuracy check against official facts or an action plan, the two parts that turn figures into work.

