Matching rules

Matching rules control how the words in an AI answer get credited to you and your competitors. When an engine calls your brand by a nickname, a product name, or a misspelling, a rule teaches MentionFlow that those all mean you — and a re-scan can apply the new rule to answers you have already collected. It lives at Matching rules.

What it does

The page has four tools:

• Competitor aliases — extra names for each competitor, so a mention under any of them is counted correctly.
• Custom patterns — an advanced rule for names that aliases can't express, like spelling variants or glued-together product names.
• Chart colors — the fixed color each brand keeps across every chart and table.
• Apply rules to history — re-run your current rules over already-collected answers.

Who it's for

• Anyone whose brand or competitors get called several things by AI assistants — abbreviations, product lines, common misspellings.
• Analysts who want their charts to read consistently, with each brand in a fixed, recognisable color.

Competitor aliases

An alias is an alternate name — a product name, shorthand, or misspelling — that should count as the same competitor. (Your own brand's aliases live on the Brand hub, not here.)

Try it:

1. Open Matching rules and find the Competitor aliases card. (If you have no competitors yet, add them on the Competitors page first.)
2. Edit a competitor row and type the alternate names, comma-separated — the field prompts you with "Product names, shorthand, misspellings — comma-separated".
3. Save. From now on, a mention under any of those names is credited to that competitor.

Aliases are matched case-insensitively, capped at 20 per competitor and 2–60 characters each. An alias that collides with another tracked brand is refused — "one alias can't attribute to two competitors" — so a name can never credit two companies at once.

Custom patterns (advanced)

Some names aliases can't capture — a brand written with or without a space or hyphen ("Acme Cloud", "acme-cloud", "acmecloud"), for example. A custom pattern handles those. It is an advanced tool, so it comes with a built-in Test button and strong safety limits.

Two things are worth knowing:

• A dot means a literal dot. Unlike normal pattern syntax, a . matches an actual full stop — handy for domains, so acme.io matches the text "acme.io" and not "acmeXio". Where you really do want "any character", use a character class like [a-z0-9].
• Patterns run in guaranteed linear time. A safe subset is allowed (no lookarounds, backreferences, or quantified groups), so a rule can never slow collection down. Patterns are case-insensitive, 2–120 characters, and capped at 12 per brand.

Try it:

1. In the Custom patterns (advanced) card, choose the brand or competitor the pattern should credit.
2. Type the pattern (the field shows an example like acme[\s-]?cloud).
3. Click Test — it checks the pattern against your 40 most recent answers and reports "Matched X of Y recent answers", with a few example snippets. It is explicit that this is a sample preview, not your full history.
4. Click Add rule to save it.

A pattern that would also match a different tracked brand is refused, so one entity's rule can never steal another's mentions. If a pattern is too broad or too complex, the page tells you exactly what to fix.

Apply rules to history

New rules apply to future answers automatically. Re-scanning applies them to answers you have already collected, too — the same matcher, run backward over your history.

Try it:

1. In the Apply rules to history card, click Re-scan history and confirm the dialog: "Re-scan collected answers with the current rules?"
2. The re-scan re-runs name, alias, and pattern matching over your stored answers and updates mentions to match — new rules add mentions, removed aliases take theirs away.
3. When it finishes you'll see a summary: how many answers were scanned, and how many mentions were added, removed, or confirmed.

A few honest details:

• AI-scored sentiment is kept; newly found mentions start unscored (an em-dash, not neutral — see Sentiment). The re-scan never re-runs the language model.
• Metrics recompute from the updated mentions, and the re-scan can't be undone.
• It works newest answers first, and if it doesn't finish in one pass it remembers where it stopped — "Run again to continue from where this run stopped." Only one re-scan runs per brand at a time.

Chart colors

Give each brand a fixed color that follows it across every chart and table. Each row offers eight preset swatches plus a free color picker, and a Default button to clear an override. Custom colors are applied as-is in both light and dark mode, so mid-brightness colors read best.

How matching decides what counts

• Your tracked name and its aliases are matched on word boundaries, case-insensitively, so a name inside a longer word doesn't falsely count.
• Custom patterns are applied exactly as written, case-insensitively, through the safe matcher.
• When several brands appear in one answer, each is credited at its earliest appearance, and mentions are numbered in the order they appear in the text.
• Cross-attribution is blocked at the source: an alias or pattern that also matches any other tracked brand is refused when you add it, so rules can't misattribute mentions.
• These matches feed the mention rows that drive your metrics — which is why adding an alias or pattern adds mentions, and a re-scan is what carries a rule change back over your history.

Limits

• Editor rights and a live workspace are required; view-only members and the demo can't edit rules.
• These are your workspace's rules only — they never touch the shared global brand registry.
• Aliases: 20 per competitor. Custom patterns: 12 per brand. Re-scan: one at a time per brand.

Related

• Competitors — where competitors are added before you alias them.
• Brand and knowledge — where your own brand's aliases live.
• Answers and receipts — where mentions and their evidence are shown.
• Sentiment — why re-scanned mentions start unscored.