Should I track branded prompts, unbranded prompts, or both?

Branded and unbranded prompts answer two different questions. Mixing them in one score makes both answers harder to read.

A branded prompt names you. "What does [your brand] charge?" or "Is [your brand] HIPAA compliant?" The engine already knows who you are. The question is whether it gets your facts right.

An unbranded prompt describes the need, not the vendor. "Best scheduling software for a two-location dental clinic" is one. The question is whether the engine names you at all.

Say you run a hypothetical scheduling tool. Engines describe your pricing perfectly when asked by name, but never mention you in category answers. A blended score averages those two results into something middling. It tells you nothing about where to act. Split them and the picture is plain: accuracy is fine, discovery is the gap.

The playbook on splitting branded and unbranded prompts walks through the mechanics of tagging each prompt by type.

The two buckets fail in different ways, and the fixes sit in different places.

Fixing branded answers

Stating a specific fact (a price, a number, a one-line definition) on a page that can be fetched gives the engine a sentence it can quote. Check that those facts appear in the HTML before JavaScript runs, because most non-Google AI crawlers do not execute JavaScript, according to Vercel's crawler study. Then remove contradictions: the same price stated differently on your pricing page and your footer is a plausible source of wrong answers.

Fixing unbranded answers

This is harder and less in your control. Start with access: per-bot robots.txt permission, firewall reachability, and index eligibility. Beyond that, when engines cite other sites instead of yours, being named on those sites is associated with appearing in the answer. No effect size is promised.

One caution on benchmarks. Brand size is the dominant confounder in citation data. Compare your unbranded inclusion against brands of similar size, or you are mostly measuring fame.

Your prompts are not the queries engines actually run. As of October 2026, the major engines document rewriting:

• OpenAI says ChatGPT search typically rewrites your query into one or more targeted queries that it sends to search providers, and saved memories can shape that rewrite.
• Google says AI Overviews and AI Mode may use a query fan-out technique, issuing multiple related searches across subtopics.
• Google's Gemini API grounding tool automatically generates one or multiple search queries and runs them.

So a prompt tracker measures answers to an input that only approximates the real retrieval input. That does not make tracking useless. It means you should treat each number as a sample, not a census.

Retrieval is also discretionary. ChatGPT decides per query whether to search at all. When it doesn't, the answer comes from model memory frozen at a knowledge cutoff. OpenAI lists a February 16, 2026 cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path. Branded prompts are especially exposed here: a stale ungrounded answer can repeat last year's pricing.

The closest view of real queries is Bing Webmaster Tools' AI Performance report, which shows grounding queries for Copilot and Bing AI summaries. You can compare those grounding queries with your prompt set to see how far your proxy drifts. For the wider trade-off, see prompt-based tracking versus engine-reported data.

Competitor-name prompts

"[Your brand] vs [competitor]" mentions you, so it is technically branded. But it also tests how you are positioned against a rival. Keep these as a third bucket. The page on tracking prompts that mention competitors by name covers when they earn their place.

Grounded versus ungrounded answers

Record whether each answer showed citations. A wrong fact in a grounded answer points to a page the engine fetched. A wrong fact in an ungrounded answer points to model memory, which nothing you publish this quarter can change.

Small samples

Accuracy rates on a handful of answers swing wildly. One flagged answer out of two looks like a collapse. Report a per-platform accuracy figure only when enough answers came back to make it meaningful.

Growing the prompt set

Adding prompts should not move your score by itself. If your tool sums penalties without dividing by answer count, a bigger plan produces a lower score for the same error rate. Whether more prompts should change your score explains the fix: normalize by answers returned.

1. Tag every prompt as branded, unbranded, or competitor-comparison. Use the prompt set template as a starting structure.
2. Write branded prompts around facts you can verify: pricing, plans, integrations, compliance, location. Keep a source-of-truth list so you can mark answers right or wrong.
3. Write unbranded prompts from buyer needs, phrased the way a customer would ask, not stuffed with your keywords.
4. Score each bucket on its own: error rate for branded, mention rate for unbranded.
5. Measure each engine separately and run logged out with memory off, so results are repeatable. The ChatGPT prompt tracking method shows how.
6. Check access before content. If a crawler is blocked, neither bucket can improve through retrieval.

For a quick first look, the free checker on this site runs up to three buyer questions and scores the answers. It is a snapshot, not ongoing tracking.

Which should I start with if I can only track one type? Start with whichever problem costs you more. If prospects arrive already knowing your name and repeat wrong pricing on calls, branded accuracy is urgent. If you are a newer brand that buyers have not heard of, unbranded inclusion matters more. Either way, add the second bucket as soon as you can, because the two can move independently.

Should my prompts match the exact words customers type? Aim for realistic wording, but do not expect a perfect match to matter much. Engines rewrite prompts into their own search queries before retrieving pages, so the exact phrasing a customer types is not what hits the index. Topical fit matters more than wording. Google also says you do not need to chase every long-tail variation for its AI features.

Why does my unbranded visibility differ between ChatGPT and Google AI Overviews? They draw on different retrieval systems and models. Google states that AI Overviews and AI Mode may use different models and techniques even from each other. ChatGPT searches at its own discretion. Index status also differs by engine: a page in Bing but not Google can be cited by Copilot while Google's AI surfaces ignore it. Score each engine separately.

Will adding schema markup get me named in unbranded answers? The evidence does not support that. Google says structured data is not required for its generative AI search features. Ahrefs' matched-control study of 1,885 pages found no major uplift in AI citations after adding schema, and a 4.6% decline on AI Overviews. Existing markup does no harm, but it is not a lever for being named.

Does ranking in Google's top 10 mean I'll appear in unbranded AI answers? Not reliably. Ahrefs found in March 2026 that 37.9% of URLs cited in AI Overviews also appeared in the first 10 results, so most citations came from elsewhere. Ranking still helps you be eligible, since AI Overviews only cite indexed pages that can show a snippet, but it is not a guarantee of inclusion.