Compare AI Answer Snapshots Fairly in Shane’s Gap Analyzer

An AI answer snapshot is most useful when you can explain what you asked, what you saw and which other checks belong beside it. A percentage without that context can hide changes in the questions, engines or brands being measured.

Shane’s Gap Analyzer gives you a local workspace for saving pasted answers, source links and content gaps. Its AI answer panel also calculates a mention percentage. Here is a practical way to use those snapshots for a fair comparison and turn the findings into a focused writing task.

1. Define the comparison before collecting answers

Choose one brand, a small set of exact prompts and the engines you will check. Record the date, market, relevant product settings and whether a prompt names your brand. Keep that capture log alongside your research so you can repeat the same process later.

For example, a fictional bookkeeping company might review three questions on two engines: choosing software for a sole trader, preparing invoices and comparing reporting features. That produces six planned observations per round. These are example research choices, not measured results or a sample-size recommendation.

Separate deliberately branded questions from open questions. “Compare Example Ledger with another tool” gives the system a brand name that “How should a sole trader choose bookkeeping software?” does not. Combining the two can make a change in your prompt mix look like a change in visibility.

The tool saves a timestamp when you add a snapshot. That timestamp records the save, not necessarily when the answer was originally generated. Keep the original capture time in your supporting research notes, especially when entering older observations.

2. Set the brand and website before saving

Enter Brand name and Your website before using Analyze and save snapshot. Use a full website address, then keep those settings consistent for the batch.

The current mention check looks for the brand text or website domain in the combined answer text and pasted citation entries, ignoring letter case. A match produces “Brand/domain mentioned.” This is a text match, so review the answer yourself before classifying its meaning.

• A brand appearing only in a pasted citation can trigger the label even when the answer does not name it.
• A sentence such as “Example Ledger is not included in this comparison” still contains the brand text.
• A short brand name can match part of a longer word. The check does not require a separate word or an endorsement.

The result is stored when the snapshot is saved. Changing the brand or website later does not recalculate the labels on older snapshots. Saved snapshots also do not each retain a separate copy of those brand and website settings. Mixing brands in one workspace can therefore leave a percentage built from checks made under different conditions.

Keep one consistent brand context for a comparison, and document it in your capture log. If you already mixed contexts, review the original answers and group them outside the tool before reporting a result. Do not assume that changing the setting has corrected the old records.

3. Keep citation entries connected to the evidence

Paste the answer you actually observed into AI answer text. Put each source URL on its own line in Cited source URLs, then use Facts / details the answer misses for your editorial observation. Avoid adding your own summary to the answer field, since that text would also participate in the mention check.

The card’s cited-source count is the number of nonempty lines you entered. It does not check that each line is a working URL, remove duplicate URLs or confirm that the AI answer cited the page. Open the original sources and check their relevance yourself. If the same URL is pasted twice, it contributes two entries.

Cards show an excerpt when an answer exceeds 500 characters. The full saved answer remains in the workspace export, and the mention check uses the full answer rather than just that visible excerpt. A brand appearing later in a long answer can explain a label that looks surprising on the card.

4. Read the percentage with its full denominator

Brand visibility checks and Visibility from your checks use all saved AI snapshots. The calculation divides the number with a stored positive mention label by the total number of saved snapshots, then rounds the result to a whole percentage. With no snapshots, the top metric shows a dash.

The engine filter narrows the cards shown under AI citation gaps. It does not narrow the dashboard calculation. Choosing ChatGPT in that filter does not turn the percentage into a ChatGPT-only result.

Consider a fictional workspace with four saved checks: two ChatGPT snapshots marked mentioned and two Gemini snapshots marked absent. The dashboard shows 50%. Filter the cards to ChatGPT and the dashboard still shows 50%, even though both visible cards have positive labels.

Repeated saves count as separate snapshots too. Saving the same answer again changes its weight in the sample; the tool does not deduplicate observations. Check for accidental repeats before using the total in a report.

5. Report a reviewed set of observations

Use Export workspace to download shanes-gap-analyzer-workspace.json. It includes the saved AI snapshots, SERP records, briefs and workspace settings, regardless of the selected engine filter. Keep an unchanged copy, then review the relevant records in your own analysis document or spreadsheet.

For each comparison, state the brand, prompt list, engine, capture period and rule for deciding whether a mention was meaningful. Separate brand presence, a recommendation and a citation to your website. They answer different questions, and the tool’s single label does not distinguish them.

A useful report might say: “In this selected set of six answers, our review found the brand named in two. One of those answers recommended it; neither linked to our website.” This is a fictional reporting example. Only use counts that your saved evidence supports.

When repeating the check, compare the same prompts and engines, and call out missing observations. A change in a small research sample does not establish a market-wide visibility rate or prove that a page edit caused the change.

6. Turn one observation into a writing task

Choose a specific saved snapshot and click its Draft content brief button. This is clearer than using Create brief from latest check when reviewing a filtered list: that shortcut uses the most recently saved AI snapshot across the workspace, which may belong to another engine.

Review the generated outline before saving or copying it. The brief builder reads its target page from the current form fields rather than automatically restoring the snapshot’s saved target URL. Check that page, the wording, source list and missing facts against your evidence. Write one bounded task, such as adding a supported explanation of an invoicing feature to a relevant page. Assign a source to verify rather than promising a future AI citation.

Records stay in the current browser. Export important research before clearing browser data, and keep the original answer evidence with it. The interface imports SERP CSV files; it does not provide a full-workspace JSON restore control. Treat the export as a portable research record, not a promise of one-click recovery.

Shane’s Gap Analyzer does not query AI engines or verify citations automatically. Its value here is keeping your observations and next writing decisions together, with enough context to explain what the numbers actually describe.