Free AI Visibility Tracking: What a One-Off Snapshot Misses

A free check runs a few buyer questions through AI engines once. It records whether your brand is named, which sites are cited, and what facts the answer states about you.

That is useful. It can show a wrong price today, or a competitor named where you expected to be.

It cannot show a trend. AI answers vary between runs, and as of October 2026 most engines decide per query whether to search the web at all. One run can't separate a real shift from noise. That is the core of whether a one-time check is enough: it answers "what does AI say now", not "is it getting better".

Tracking is the same check repeated with fixed inputs, so the only thing that changes between runs is the engine.

You need a spreadsheet, a calendar reminder and discipline about inputs.

1. Write a fixed prompt set. Pick 5 to 15 questions a buyer would ask before choosing a product like yours. Follow how to build a prompt set to mix category, comparison and problem questions.
2. Freeze the wording. Give each prompt an ID. If you change a prompt, it becomes a new prompt with a new ID.
3. Use a clean session. Run logged out, or with memory and custom instructions off. OpenAI says ChatGPT may use saved memories when rewriting a search query, so a personalised session is not reproducible.
4. Run every prompt on every engine on the same day. ChatGPT, Perplexity, Gemini, Claude and a Google search for AI Overviews.
5. Log each answer as one row. Save the raw text or a screenshot next to it.
6. Repeat on a fixed cadence. Monthly suits most teams. See how often to re-run checks.

What each row should hold

Add one access check per run

Answers depend on crawlers reaching your site. A quick test from your terminal:

curl -s https://example.com/robots.txt
curl -I -A "OAI-SearchBot" https://example.com/pricing
curl -I -A "PerplexityBot" https://example.com/pricing

A 403 or a challenge page is worth investigating. Real crawlers send the full user-agent strings listed on each operator's crawler page, and some firewalls verify IP addresses, so a laptop test is a first look, not proof.

Read it engine by engine. A blended number hides the case where one engine names you often and another never does, which is why per-platform scores beat one score.

Look for three things across runs:

• Mention changes on the same prompt. One flip is noise. The same flip across several runs is a pattern.
• Wrong facts that persist. A wrong price that repeats every month is a fix target.
• Answers with no citations. These may come from the model's training data rather than a live search. OpenAI's model docs list a February 16, 2026 knowledge cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path, so an ungrounded answer can describe an older version of your business.

Then the next move. If a fact is wrong, check that the correct fact appears in your page's HTML before JavaScript runs: Vercel found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. If engines cite third-party sites instead of yours, being named on those sites is associated with appearing in the answer, though nothing guarantees it.

Any prompt-based tracking, free or paid, measures a proxy. Be plain with your team about four gaps.

• Real user prompts. OpenAI says ChatGPT search rewrites a question into one or more targeted queries. You test your wording, not the queries the engine actually ran.
• Retrieval state. Engines decide per query whether to search. Visible citations suggest a search happened. Their absence tells you little for certain.
• Positions. As of October 2026, Google's Search Console report for AI features shows impressions only, with no position. No outside tool has ground truth for an "AI rank".
• Engine-reported totals. Bing Webmaster Tools shows grounding queries and citation counts in its interface. In Otterly's 2,000-blog experiment, Microsoft's report counted 42,817 citations for one site against 3,934 logged by Otterly's prompt tracker. The windows may differ, but a prompt sample is a sample.

The practical answer is to pair your log with engine-reported data where it exists. Verify Search Console and Bing Webmaster Tools if you haven't.

The free checker on this site runs up to three buyer questions once and scores the answers. Use it as your first dated snapshot, then log it as run one.

For repeated runs without a spreadsheet, 99Visibility's Free plan covers 5 keywords, 1 audit a month and 1 brand across 2 platforms, and you can start it with no card. That is enough when:

• you track one brand,
• a handful of buyer questions covers your category,
• two engines matter most to your buyers, and
• a monthly baseline answers your question.

It is not enough if you need all five major platforms, dozens of prompts, several brands or weekly runs. Then keep the manual log for the gaps, or move to a larger setup. If ChatGPT is your only concern, tracking ChatGPT mentions for free walks through that engine alone.