Auditing Your Current AI Visibility Before Buying Any Tool

Buyer's guide

You cannot judge a tool's impact without a starting line. Here is how founders baseline AI visibility — prompt sets, citation tracking, and the numbers worth recording before any contract is signed.

• A baseline is a dated snapshot of how AI assistants answer your buyer's real questions — not a vanity score.
• Build a fixed prompt set before you evaluate any vendor, so results stay comparable over time.
• Track three things per prompt: mention, citation, and position relative to competitors.
• Record raw answer text and cited URLs; screenshots and dates protect you from revisionist reporting.
• A tool is only worth buying if it can be measured against your baseline with the same prompts and the same method.

Why the starting line matters more than the tool

Most founders evaluate AI search visibility tools the way they evaluate a new hire: on the strength of the pitch. The demo is polished, the dashboard is dense, and the promise is a lift in visibility. What is missing is the only thing that makes a lift meaningful — a number you recorded before the tool existed in your stack.

AI assistants do not rank pages the way a classic search engine does. They compose an answer, and your brand either appears inside that answer or it does not. That makes visibility binary at the prompt level and statistical at the set level. Without a baseline, you have no way to distinguish a real change in how ChatGPT or Perplexity describes you from normal variation between sessions.

How do I baseline my AI visibility?

A baseline audit is a repeatable procedure, not a one-off curiosity. Run it once before you buy, then re-run it on the same prompts at a fixed interval. The method below is deliberately manual and cheap; its purpose is to give you an independent reference point that no vendor controls.

Step 1 — Build a prompt set that mirrors real buying questions

Write down the questions your prospects actually type when they are close to a decision. Aim for breadth across the buying journey rather than volume: category questions, comparison questions, problem-first questions, and brand-adjacent questions. Twenty to forty prompts is usually enough to see a pattern without turning the audit into a project.

• Category prompts: "best [category] tools for [use case]" — tests whether you appear at all.
• Comparison prompts: "[competitor] vs [competitor]" — tests whether you are part of the considered set.
• Problem prompts: "how do I solve [specific problem]" — tests whether your content is the source of the answer.
• Brand prompts: "what is [your brand]" — tests whether the assistant describes you accurately.
• Local or niche prompts if geography or vertical matters to your buyers.

Step 2 — Run the same prompts across assistants

ChatGPT and Perplexity behave differently, and that difference is the point. Perplexity tends to surface cited sources prominently; ChatGPT often answers without an explicit citation trail. Run every prompt in both, in a fresh session, and log the date. Do not personalize the session or carry context from a previous question, because that changes the answer.

Step 3 — Record mention, citation, and position

For each prompt and each assistant, capture three signals. Mention: is your brand named in the answer at all? Citation: is a URL you own listed as a source? Position: where do you appear relative to named competitors — first, in the middle of a list, or absent? These three signals are what any credible tool should be able to move, and they are what you will compare against later.

Step 4 — Store the raw evidence

Save the verbatim answer text and the cited URLs, not just a score. AI answers change as models are updated, and a vendor report that shows improvement is only meaningful if you can compare it to the exact wording you captured earlier. A dated document with pasted answers is unglamorous and completely sufficient.

What should I measure before buying an AI SEO tool?

Before you evaluate any vendor, decide which metrics you will hold them to. The list below separates signals you can verify yourself from signals that depend on the vendor's own reporting. Weight the first group more heavily.

1. Presence rate — the share of your prompt set where your brand is mentioned at all.
2. Citation rate — the share of prompts where a URL you own is cited as a source.
3. Competitive position — how often you appear ahead of, alongside, or behind named competitors.
4. Description accuracy — whether assistants describe your product correctly when asked directly.
5. Source quality — whether the pages being cited are ones you would want a buyer to read.
6. Content velocity — how many genuinely useful pages you can publish per month, since AI answers draw on published material.
7. Attribution path — whether AI-referred visitors can be identified in your analytics at all.

Notice what is absent from that list: a single composite "visibility score." Composite scores are convenient for dashboards and useless for auditing, because you cannot reconstruct how they were calculated. Insist on the underlying prompt-level data.

Red flags in a vendor's baseline claim

Once you have your own baseline, vendor claims become testable. A few patterns are worth watching for during evaluation, because they make comparison impossible even when the product itself is decent.

• Prompts change between reports, so improvement cannot be attributed to the work.
• Only aggregate scores are shown, with no prompt-level detail you can re-run yourself.
• Results are reported for one assistant only, usually the most favorable one.
• The baseline is generated after onboarding, so the vendor defines the starting point.
• Citations are counted without distinguishing your own pages from third-party mentions.

Where a content engine fits after the audit

A baseline tells you where you stand; it does not fix anything. The lever that moves AI visibility is published content that answers the prompts in your set — pages that assistants can cite when they compose an answer. That is the part most founders underestimate, because it is production work rather than analysis.

How to sequence the audit and the purchase

Run the baseline first, in a single sitting, and keep it somewhere you will find it in ninety days. Then evaluate tools against that reference. When you do buy, re-run the identical prompt set on the same schedule and compare prompt by prompt rather than score by score. If a tool cannot survive that comparison, it is not measuring visibility — it is measuring its own activity.

FAQ

How long does a baseline audit take?

A single sitting is usually enough. Build a prompt set of twenty to forty buyer questions, run each one in ChatGPT and Perplexity in fresh sessions, and paste the answers and cited URLs into a dated document. The work is repetitive rather than difficult, and it gives you an independent reference point that no vendor controls.

Can I just use a free visibility checker instead?

You can, but treat it as a convenience rather than a baseline. Free checkers rarely expose the exact prompt, the verbatim answer, and the date together, which are the three things you need to defend a comparison later. If a checker cannot show you its raw data, run the prompts manually for your core set.

What if my brand is never mentioned in AI answers today?

That is a valid and useful baseline — arguably the most useful one. A zero presence rate tells you the problem is content coverage rather than optimization, and it sets a clear target for the next audit. Record the absence explicitly, with dates and prompts, so future improvement is provable.

How often should I re-run the baseline?

Re-run on a fixed schedule, such as monthly or quarterly, using the identical prompt set. Consistency matters more than frequency: changing prompts between runs destroys comparability. Keep the old results so you can see movement prompt by prompt rather than relying on a single aggregate number.