Last month, a prospect asked ChatGPT about your company. They got an answer.
Confident, fluent, well organized. It described what you do, who you serve, roughly what you charge, and how you compare with 2 competitors you lose deals to every quarter.
Then they built a shortlist. You never saw the question, the answer, or the moment you made the cut or did not.
That conversation is running constantly, and almost nobody on your team has read the script. Most SaaS marketers check their own brand name once, see something broadly flattering, and move on.
That’s not an audit. An audit means asking the questions that actually get asked; pricing, alternatives, integrations, security, complaints and grading the answers against what is true.
It takes about 90 minutes, costs nothing, and reliably surfaces at least one thing that is quietly costing deals. A deprecated feature still being recommended. A price point that has not been accurate since 2024. A competitor positioned as the obvious choice for your best segment.
These 11 prompts are the audit. Run them in order, record the answers, and treat the output as a baseline you can re-run next quarter. What follows each prompt is what it actually tests, how to read the response, and who owns the fix when it comes back wrong.
How to Run This Audit Properly
The method takes precedence over the prompts because a sloppy run produces results you cannot act on or compare later.
Start from a clean session. Log out or use a temporary chat with memory and personalization disabled. Your own account has months of context about your company, which is precisely the context a prospect does not have.
Run each prompt in a separate conversation. Assistants carry context forward, so the fourth answer in a thread is shaped by the first three. Fresh chat per prompt, every time.
Enable web search where it is optional. Most buyers researching a vendor will be using a mode that browses. Test what they see, not what the base model recalls.
Record everything. Save the full response, the date, the model version, and the cited sources. A screenshot is not a record; you need the text to compare against next quarter and the sources to diagnose anything wrong.
Repeat the important ones. Answers vary between runs. Run your 3 most commercially important prompts 3 times each and note whether the substance holds. Having the same error 3 times is a finding; one bad answer is unimportant.
Vary the region if you sell internationally. Results differ by locale, and a description that is accurate in the US can be wrong or absent in EMEA.
Ask a colleague to grade the answers. Founders and product marketers read these responses generously, filling in context a buyer would not have. Someone outside the team spots the gap between what the answer says and what it implies, which is usually where the commercial damage sits.
This is the same discipline Growth-onomics applies in an AI readiness audit: a fixed prompt set, clean sessions, documented outputs because without it you cannot tell whether anything you changed afterwards actually worked.
Quick Comparison
| # | Prompt Tests | Red Flag |
| 1 | Baseline description | Wrong category or merged with another company |
| 2 | Segment fit | Recommends you for the wrong buyer |
| 3 | Pricing accuracy | Outdated figures or “contact sales” only |
| 4 | Unbranded category presence | You are not mentioned at all |
| 5 | Competitor displacement | Absent from the alternatives list |
| 6 | Head-to-head framing | Competitor’s positioning repeated as fact |
| 7 | Weakness narrative | Complaints you resolved a year ago |
| 8 | Integration capability | Integration denied or misdescribed |
| 9 | Security and compliance | Certifications unknown or wrong |
| 10 | Entity and company facts | Confused with a similarly named company |
| 11 | Citation sources | Every source is third-party and stale |
1. The Baseline Description
What is [Company]? What does the product do and who is it for?
What It Tests
Whether the model has an accurate, current, well-formed picture of your company at all. Everything else in this audit depends on this answer being right.
How to Read It
Check 4 things: the category it places you in, the primary use case it names, the customer type it describes, and whether any detail belongs to a different company. Compare against your own positioning statement word for word.
The Fix
Category errors usually trace to inconsistent naming across your site and third-party profiles. Standardize the description everywhere it appears, implement Organization markup with a sameAs array pointing to verified profiles, and update your review platform listings.
2. The Fit Test
Is [Company] a good choice for a [company size] [industry] team that needs [core job to be done]?
What It Tests
Whether the model understands your ideal customer profile (ICP), or whether it is guessing from generic category assumptions.
How to Read It
The failure mode here is subtle. The answer may be positive but aimed at the wrong buyer, describing you as suited to enterprises when you sell to mid-market, or to marketing teams when you sell to operations.
A confident recommendation to the wrong segment produces unqualified demos.
The Fix
Publish explicit use-case and audience pages naming the company sizes, industries, and roles you serve. Vague “for teams of all sizes” positioning gives a model nothing to work with, so it invents a fit.
3. The Pricing Question
How much does [Company] cost? What are the pricing tiers and what drives the total cost?
What It Tests
Whether accurate pricing information exists anywhere the model can reach, and what it uses when yours is unavailable.
How to Read It
Watch for figures you no longer charge, tier names you have retired, and estimates attributed to third-party directories. If the answer is “pricing is not publicly available, contact sales,” note which sources it consulted before concluding that.
The Fix
Publish the pricing structure in crawlable text even if you cannot publish figures: what drives cost, what each tier includes, and the adjacent expenses buyers ask about. Then update the marketplace and directory listings carrying old numbers.
4. The Unbranded Category Query
What are the best tools for [the job your product does] for a [your ICP] team?
What It Tests
The most commercially important prompt in this audit. It asks whether you exist in the conversation when nobody mentions your name, which is how most discovery actually happens.
How to Read It
Record whether you appear at all, your position in the list, how you are characterized, and who else appears. Run this several times, because list composition varies more than any other answer type.
The Fix
Absence here is a content and authority problem rather than a technical one. It usually means the roundups, review platforms, and community threads that feed these answers do not include you prominently, which is off-domain work. Run the source audit prompt immediately afterwards, since the domains named there become your outreach list.
5. The Competitor Displacement Test
What are the best alternatives to [your largest competitor]?
What It Tests
Whether you appear at the exact moment a buyer is looking to leave a competitor, the highest-intent query in most SaaS categories.
How to Read It
Note whether you are listed, how you are differentiated, and whether the model names a specific reason to switch to you. “Also worth considering” is much weaker than “better suited for teams that need X.”
The Fix
Build a genuinely useful alternatives page for that competitor, covering migration realities honestly. Then check whether third-party alternatives roundups include you, since those are frequently cited alongside vendor pages.
6. The Head-to-Head
[Company] vs [Competitor]: which should I choose and why?
What It Tests
Whose framing wins. Comparison answers are heavily shaped by whoever published the most structured comparison content, and that is often your competitor.
How to Read It
Look for their language appearing as neutral fact, their chosen criteria, their characterization of your limitations, their pricing framing. Also check whether the trade-offs described are actually true today.
The Fix
Publish your own comparison on the criteria buyers weigh, in a table, stating honestly where the competitor is stronger. One-sided pages get discounted, and a page that never acknowledges a trade-off reads as marketing to both models and buyers. Check the date on your existing comparison pages too, because competitors ship features and stale comparisons quietly become inaccurate ones.
7. The Complaints Prompt
What do users complain about with [Company]? What are its main limitations?
What It Tests
The weakness narrative attached to your brand, and where it comes from. This is the prompt most teams avoid and the one that most often produces something actionable.
How to Read It
Separate 3 categories: complaints that are current and fair, complaints you resolved months ago, and complaints that belong to a different product. The second category is the highest-value finding in this entire audit.
The Fix
Resolved complaints usually persist because a review left 2 years ago or an old forum thread still carries weight. Respond publicly to those reviews noting the change, publish a changelog entry naming the fix, and correct the thread with evidence.
Growth-onomics performs these types of source corrections as part of its brand mention work, because the fix is rarely on the vendor’s own site.
8. The Integration Check
Does [Company] integrate with [critical tool in your stack]? What does the integration do and how is it set up?
What It Tests
Whether the model can confirm capability. Integration questions gate deals, and a false negative here removes you from consideration silently.
How to Read It
Check for outright denial of an integration you offer, vague hedging, and misdescription of what the integration truly does. Test your 3 most commonly asked-about integrations separately.
The Fix
Give each significant integration its own page describing what it enables, what setup requires, and which plan includes it. Then audit your listings in the partner’s marketplace, which is often cited more readily than your own site.
9. The Procurement Screen
Is [Company] suitable for enterprise procurement? What certifications, data residency options, and security practices does it have?
What It Tests
Whether compliance facts are reachable. Security reviewers increasingly run a first pass through an assistant before the questionnaire arrives.
How to Read It
Look for “no information available,” outdated certification status, or hedging that implies risk. Absence reads as a negative signal to a reviewer, not as neutral.
The Fix
Publish certifications, data residency options, retention policies, and sub-processor lists as crawlable text with dates, keeping the full documentation in the trust portal.
A gated trust center is invisible to every system answering this question.
10. The Entity Test
Who founded [Company], where is it based, how large is it, and who are its investors?
What It Tests
Entity resolution: whether the model has your company cleanly separated from similarly named organizations.
How to Read It
Watch for details belonging to another business, wrong headquarters, outdated leadership, or a merged profile combining two companies. This failure is more common than teams expect, particularly for generic or shared names.
The Fix
Organization and Person schema with a complete sameAs array, consistent company descriptions across LinkedIn, Crunchbase, and review platforms, and current leadership information everywhere. Entity confusion resolves slowly, so start early.
11. The Source Audit
Which sources did you use to answer that? List the specific pages.
What It Tests
Run this as a follow-up to your worst-performing prompt above. It converts a vague complaint into a specific work list by revealing which domains shape the answer.
How to Read It
Classify each source: your own pages, shared properties like review profiles and marketplace listings, and independent sources. The mix tells you where the fix belongs. Note that assistants do not always report sources reliably, so treat the list as a strong hint rather than a complete record.
The Fix
Whatever the sources say. If review profiles dominate, update them. If a single outdated roundup is doing the damage, pitch the publisher a correction. If your own pages are absent entirely, the problem is technical or structural rather than reputational.
What to Do With the Results
90 minutes of prompting produces a list of problems with 3 different owners, and sorting them correctly is what turns an audit into progress.
Technical fixes come first. If your own pages are absent from source lists entirely, check indexation, snippet directives, and whether critical content renders without JavaScript. No amount of content work helps a page that cannot be used.
Content fixes come second. Missing pricing explanations, absent integration pages, no comparison content, gated security documentation. These are the gaps where the model wanted an answer from you and had to go elsewhere.
Off-domain corrections come third and take the longest. Stale reviews, outdated roundups, uncorrected community threads. These compound slowly and cannot be rushed, which is why they should start immediately even though they finish last.
Then set the baseline properly. Save the full text of every answer with the date, freeze the prompt set, and re-run it quarterly. The value is in the comparison, not the first snapshot; a description that improves over two quarters is evidence, while a single good answer is an anecdote. Models update, sources shift, and a finding from March may have quietly reversed by June, so the schedule matters as much as the first run.
Conclusion
The strange thing about AI-assisted buying is how much of it happens in private. There’s no impression count for a bad answer, no lost-click report when an assistant recommends a competitor, no alert when a model starts quoting a price you retired 18 months ago. The only way to know is to ask.
These eleven prompts are the cheapest diagnostic available in AI visibility work. They require no tooling, no budget approval, and no vendor. What they need is the discipline to run them cleanly, record the results, and read the answers as a skeptical buyer rather than as a proud founder.
Most teams running this for the first time find the same result: the basic description is broadly right, the commercially pertinent details are wrong or missing, and the unbranded category query is actually the problem. That pattern is fixable, and the order of work is knowable once you have the source lists.
If you would rather have someone run this properly across platforms, with a documented prompt set and a plan attached to the findings, the Growth-onomics team can handle the audit and show you exactly where your answers are coming from.
FAQs
How often should I run this audit?
Quarterly for the full set, monthly for your 3 most commercially important prompts, typically the unbranded category query, the pricing question, and the competitor displacement test. Models update, sources change, and answers drift without any action on your part. Running the same frozen prompt set each time is what makes the comparison meaningful; changing prompts and measuring improvement simultaneously produces a number you cannot interpret. Add an ad-hoc run after any major product launch, rebrand, or pricing change, since those are the moments when descriptions most reliably go stale.
Should I test other assistants too?
Yes, and the results will differ enough to be worth the extra time. ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews use different retrieval approaches and lean on different sources, so a brand can be well represented in one and invisible in another. Start with ChatGPT because of its scale, then add whichever platform your buyers actually mention in sales calls. Run the same prompts across each, record results separately, and never average them into a single score; the divergence between platforms is itself a useful finding.
What if the answers are different every time I run the prompt?
That is expected, and it is why repetition is essential. Assistants sample probabilistically, so phrasing, timing, and session context all shift results. Run your most important prompts at least 3 times and look at what stays constant. If your product is described accurately twice and wrongly once, note the error but treat it as low priority. If the same wrong detail appears every time, that is a stable finding with a source you can trace and fix.
Can I just use a tool instead of prompting manually?
Tools are better for tracking at scale and worse for understanding. A platform running hundreds of prompts weekly gives you trend data no manual process can match, which is the right instrument for ongoing measurement. But reading full answers yourself surfaces things a visibility score never will: the tone of a recommendation, a subtly wrong framing, a competitor’s language repeated as fact. The practical combination is manual auditing to understand and diagnose, tooling to monitor and report.
Who inside the company should run this?
Whoever will act on the results, which usually means product marketing rather than SEO. The findings cut across positioning, pricing communication, competitive framing, and documentation, and most of the fixes sit outside a search team’s remit. Bring in technical SEO for the indexation and rendering questions, support or customer success to validate the complaint findings, and sales to confirm whether the fit test matches what prospects actually ask. The audit is cheap enough that the real constraint is ownership, not effort.
