Before you report how often competitors appear in AI answers, check what your log is actually counting. A repeated name, a space-only entry between commas or two spellings of the same product can turn a useful observation into a misleading summary.
Shane’s AI Visibility Tracker gives you a place to record those observations manually. Its Competitor mentions counter adds the entries in the Competitors mentioned field across saved checks. It does not identify companies, merge aliases or read the original answers. A short cleanup routine makes that total easier to interpret.
Understand the unit behind the counter
The current tool splits each saved competitor field at every comma, ignores completely empty pieces and counts the remaining pieces. It then adds those counts across the workspace. Each surviving piece contributes one, even if another piece has exactly the same name.
Northstar,Harborcontributes two.Northstar,Northstaralso contributes two. The tool does not remove the repeated name.Northstar,,Harborcontributes two because the middle piece is empty.Northstar, ,Harborcontributes three because the middle piece contains a space.
Spaces around the whole field are trimmed when you save, but the individual comma-separated pieces are not trimmed before counting. An entirely blank field contributes zero. Quotation marks do not provide a way to protect a comma inside a company name: the counter still splits at that comma.
This means the dashboard is a count of entered pieces, not a count of unique competitors or every time a name appears in an answer. Decide what each piece should represent before building a report.
Use one competitor entry per reviewed answer
A useful convention is to list each competitor once for each answer you actually reviewed. If Northstar appears in the opening paragraph and again in a comparison, enter Northstar once. If Northstar appears in two separate reviewed answers, keep it in both records. The second observation is meaningful; the repeated name within one record usually is not.
Keep one saved check for each observed answer, with its exact prompt, platform and check date. Intentional repeat checks can stay separate when they represent different answers. Saving the same answer twice by mistake increases both the number of checks and its contribution to the competitor total.
A blank competitor field should have a clear meaning in your research process. After reviewing a complete answer, it can mean that none of the competitors you are tracking appeared. If the answer was unavailable or the review is unfinished, preserve that limitation in your research notes and exclude it from any completed-answer denominator.
Choose names that stay consistent
Keep a small naming key beside your research files. For each competitor, record the preferred label and any verified variations you have seen in answers. For example, you might map “Northstar Ledger” and a confirmed abbreviation to one label, while keeping the original answer in your evidence archive.
Check the identity before merging names. A parent company and one of its products may answer different research questions. Two products with similar names may belong to different businesses. If you cannot establish a match, leave the observation unresolved rather than inflating your certainty.
Use comma-free labels in the tracker. For a company whose full name contains a comma, use a clear short label and preserve its full name in the naming key. Separate the labels with commas and remove accidental empty or space-only entries before saving.
Apply the same discipline to AI platform. The tracker trims outer whitespace, but ChatGPT and chatgpt count as two labels under Models tracked. Keep product naming consistent, and record meaningful model or mode differences in your supporting evidence.
Work through a three-answer example
The following is fictional data used to explain the counter. Imagine that you reviewed three answers and saved these competitor fields:
- Answer A:
Northstar,Northstar - Answer B:
Harbor, ,Northstar - Answer C: a blank field after confirming that neither competitor appeared.
The dashboard shows 5 Competitor mentions: two from A, three from B and zero from C. That does not mean five different companies appeared. One contribution is a duplicate name and another is only a space.
Under the one-entry-per-competitor-per-answer convention, correct A to Northstar and B to Harbor,Northstar. Leave C blank. The cleaned total is 3, covering 2 distinct competitors across 3 reviewed answers.
A clearer report would say: “In these three reviewed answers, Northstar appeared in two and Harbor appeared in one. Two distinct competitors appeared overall.” Those figures answer different questions. Neither establishes how often all users see those companies or whether an answer recommends them.
Correct saved entries without losing the original check
The current log has a Remove control but no edit control for a saved row. Before making corrections, export the log and keep the original answer evidence. Then use a careful replacement process:
- Copy the original prompt, platform, brand, check date and source notes into your working notes.
- Create the corrected record with the intended competitor labels and the original check date. Confirm its answer-type selection against your evidence as well.
- Inspect the new row, then remove only the superseded record. Until you remove it, both versions contribute to the totals.
- Check that the number of saved checks returns to the expected total and that the corrected competitor field is visible.
Do not delete legitimate repeat checks simply because their prompts match. Use the date and supporting evidence to distinguish a correction from a separate observation. Keep your cleanup decision with the exported analysis so another reviewer can follow it.
Calculate the report from the intended records
The log’s search field can help locate a prompt, brand, competitor or platform. It narrows the displayed rows, but the dashboard totals and Export CSV still cover all saved records. The current search does not search the date or source-notes fields.
Export the CSV, keep an unchanged copy, and make a working copy for the reporting period and prompt set you want. Count distinct competitor labels separately from the number of reviewed answers containing each one. Review incomplete checks and accidental duplicates before calculating percentages.
Record your selection rule alongside the result: which prompts, dates, platforms and completed answers were included. Multiple competitors can appear in the same answer, so competitor appearance rates need not add to 100%. The existing evidence-notes guide explains how to retain the answer context and review the separate brand-mention indicator.
The practical goal is a report that someone can reconstruct: consistent labels, one clearly defined observation per record, and a total whose meaning is stated. The tracker organizes your manual checks; your review supplies the interpretation.