The problem with “we showed up when we searched ourselves”
Most leadership teams run the same informal test. Someone types the company name into ChatGPT, likes the answer, and declares the brand safe.
That test proves almost nothing. Buyers who already know your name are not your risk. Your risk is the CFO who asks an assistant to compare three vendors in your category, or the CTO who asks which platforms handle a specific integration. The answer names a short list. If you’re missing, there is no rejection email and no lost-deal note. The deal simply never appears in your pipeline.
Before you start, accept three constraints:
- Answers vary. The same prompt can produce different results across runs, platforms, and dates. Measure patterns, not single outputs.
- There is no official AI ranking. Every metric here is directional. That’s still more useful than anecdotes.
- Your website is one input among many. Review sites, forums, analyst pages, and directories often weigh heavily in what models say about you.
Step 1: Rank questions by revenue, not by volume
Traditional keyword research ranks queries by search volume. AI audits should rank questions by commercial consequence.
Start with 50 to 80 questions and sort them into four buckets:
| Bucket | Example question | Revenue weight |
|---|---|---|
| Shortlist questions | “Which vendors should a mid-market company consider for [category]?” | Highest |
| Comparison questions | “[Brand A] vs [Brand B]: which fits a regulated industry?” | High |
| Risk questions | “What are the downsides of [your brand]? Any security or integration concerns?” | High |
| Education questions | “How do we reduce [problem] without replacing our current stack?” | Medium (builds early familiarity) |
Source the wording from places your market actually speaks: sales call recordings, RFP language, support tickets, and objections your reps hear weekly. Tag each question by buyer role (CEO, CFO, CTO), product line, and deal stage. Weight shortlist and comparison questions most heavily, since they sit closest to a buying decision.
Step 2: Capture a repeatable baseline
A baseline is only useful if someone can reproduce it next month. Standardize the process before you collect anything.
- Choose platforms by buyer behavior. Cover ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Microsoft Copilot, and Claude. Ask your sales team which ones prospects mention.
- Run every question multiple times. Three runs per platform is a sensible minimum to dampen randomness.
- Isolate each session. Use fresh profiles with memory and personalization off.
- Log the context. Record platform, model version, date, region, and whether live web search was on.
- Store the raw output. Keep the full answer, the order brands appear in, and every cited link, so you can re-grade later without re-running.
A shared spreadsheet is fine for a first pass. Once you pass a few hundred prompt runs, automation pays for itself.
Step 3: Grade each answer with a rubric
Presence alone is a shallow measure. An answer can name you and still lose you the deal. Grade every answer on five dimensions:
| Dimension | Question to ask | Scoring approach |
|---|---|---|
| Mention | Is the brand named at all? | Yes or no |
| Position | Is it recommended first, listed among others, or buried? | Top, middle, or tail |
| Citation | Does the answer link to your domain? | Yes or no, plus which page |
| Accuracy | Are pricing, features, integrations, and company facts correct? | Correct, outdated, or wrong |
| Framing | Are you the leader, a viable alternative, a niche option, or a caution? | Four-level tag |
Then roll the results up into a one-page view. The numbers below are illustrative only, to show the format:
| Platform | Mention rate | Citation rate | Accuracy issues | Dominant framing |
|---|---|---|---|---|
| Platform A | 40% | 15% | 2 outdated facts | Alternative |
| Platform B | 25% | 5% | 1 wrong fact | Niche |
| Platform C | 55% | 30% | None found | Leader |
Look for gaps between rows. A brand that is strong on one platform and invisible on another usually has a source-layer problem, not a brand problem. Separate the accuracy column for your executive audience. Wrong pricing, discontinued products, and false claims about security or compliance do real commercial damage, and they are usually the fastest fixes.
Step 4: Trace the evidence behind the answers
You can’t improve what you can’t explain. For every question where a competitor appears and you don’t, find out what the model leaned on. Four checks cover most cases.
Source mapping. List every domain cited in those answers: review platforms, trade publications, comparison articles, community threads, documentation. Count how often each appears. The top of that list is your off-site priority list, and it often looks nothing like your current PR plan.
Crawl access. Confirm that your robots.txt isn’t blocking search and AI crawlers you actually want, that key pages load their content without depending on heavy client-side scripts, and that pages are fast and structurally clean. Many B2B visibility gaps come from engineering decisions made years ago for other reasons.
Entity consistency. Compare how your company is described on your site, LinkedIn, Crunchbase, review profiles, and partner directories. Conflicting category labels, old descriptions, or mismatched names make it harder for a model to associate you with a topic. Apply accurate Organization, Product, and FAQ schema to reinforce the picture.
Answer-ready content. Check your priority pages for a direct answer in the opening lines, descriptive headings, comparison tables, and specific proof such as named outcomes and original data. Pages built around vague positioning language give a model little to quote.
Step 5: Assign owners, fix, and report on a schedule
An audit that ends in a slide deck changes nothing. Convert findings into an owned backlog.
| Finding | Owner | Fix | Typical timeline |
|---|---|---|---|
| Wrong or outdated facts | Marketing ops | Update site, profiles, directories at the source | 1–3 weeks |
| Crawl or rendering blockers | Engineering | Fix access, rendering, schema, speed | 2–6 weeks |
| Weak comparison and answer pages | Content and product marketing | Rebuild with direct answers, tables, proof | 4–8 weeks |
| Missing from cited third-party sources | PR, partnerships, customer marketing | Earn reviews, listings, features, contributed expertise | 2–6 months |
| No original reference material | Content and research | Publish data, benchmarks, and methodology others cite | Ongoing |
Timelines are planning guides, and results depend on competition and your starting authority.
Next, connect visibility to revenue, because finance will ask:
- Create a GA4 channel group for referrals from AI platforms.
- Add a free-text “How did you hear about us?” field to demo and contact forms.
- Have sales log any AI mention in discovery calls.
- Watch branded search and direct traffic for lift, since AI-influenced visits often arrive that way.
Repeat the audit monthly with the same core question set so trends are comparable, and refresh the list each quarter.
Four mistakes that distort the results
- Averaging across all questions. A strong score on education questions can hide absence on shortlist questions.
- Skipping the competitor lens. Your score means little without the same grading applied to two or three rivals.
- Fixing content only. If crawlers can’t read your site, better copy won’t help.
- Chasing shortcuts. Mass-produced pages and manipulated mentions are brittle. Durable visibility comes from accurate facts, clear identity, and credible sources.
What this means for your business
AI assistants now sit between buyers and your website. They shape the shortlist before your analytics register a visit. The audit above gives you a baseline, a diagnosis, and an owned plan, and it costs far less than one lost enterprise deal.
At Techusu, strategists and engineers work in the same room on this: the content models should cite, the technical foundation that lets them read you, and the tracking that links visibility to pipeline. To see where your brand stands across AI platforms, start at techusu.com.
Frequently asked questions
1. What does it mean to audit your brand in AI answers?
2. How many questions should a B2B company test?
3. Which AI platforms should be included?
4. Why does an AI assistant get facts about my company wrong?
5. How do we measure revenue impact from AI visibility?
Simran
Content WriterSimran writes practical guides on search, tracking, site performance and software for B2B companies at Techusu.
Related reading
B2BAccount-Based Advertising: A Realistic Guide for Mid-Market B2B
Simran · 20/11/20267 views
AEO vs GEO vs SEO: What B2B Leaders Should Fund in 2027
Simran · 28/10/2026239 views
AI Workflow AutomationAI Workflow Automation: 6 B2B Processes Worth Automating First (and 3 That Aren’t)
Simran · 17/10/2026318 viewsNever miss an insight
One email a month on growth, search and technology, written for decision makers.
