N-Gram Analyzer

A contiguous run of words from a search term. "Stainless steel water bottle" contains the unigrams stainless, steel, water, bottle; the bigrams stainless steel, steel water, water bottle; and so on. The analysis gives each of those the full metrics of the term it came from, then adds up every appearance — so a word that cost you $2 in each of 200 terms shows up as $400 in a single row.

Unigrams. They give the broadest view and the biggest wins, because one word blocks every phrase containing it. They also carry the highest false-positive risk, which is exactly what the converting-terms column and the drill-down are for. Move to bigrams when a word looks wasteful in aggregate but you can see it belongs to your category — the two-word phrase gives you the context to block precisely. Trigrams are mostly for confirming a pattern rather than acting on one.

Because the deciding column is relevance, and relevance is not in your data. The tool can prove that "steel" appears in fourteen search terms, cost you $180 and never converted. It cannot know whether you sell steel products. A tool that skips that step and hands you an export is guessing on your behalf with your traffic — so this one proposes, shows its working, and waits for you to tick the box.

When a word has converting terms behind it, blocking the whole word costs you those sales. So the tool looks only at the non-converting terms containing that word, finds the two-word phrase carrying the most of that waste, and then checks that phrase against every converting term to confirm it appears in none of them. If one qualifies, you get a targeted negative that removes the waste and leaves the working traffic alone.

Negative exact blocks precisely one search term — useful for surgical removals, useless for a pattern. Negative phrase blocks any search term containing your words in order, which is exactly what word-level analysis is for: one entry removes every current variation and every future one. Note that Amazon does not support negative broad match at all, despite how often people assume it does — phrase and exact are your only options.

Because they appear in a large share of all search terms, so they accumulate enormous spend and look like your biggest waste candidates — and blocking one would remove most of your traffic overnight. They carry no information about relevance. The toggle is left visible rather than the filter being silent, so you can see the decision was made and reverse it if you want to inspect them.

Because an ASIN is not language. Breaking b0dd8bstkd into words produces a meaningless token that would sit in your counts as though it were a real pattern. They are excluded, and the count and spend are shown separately so you know how much was set aside. If ASIN traffic is wasting money, the fix is a negative product target rather than a negative keyword — a different entity, handled in the Keyword Harvester.

Sixty to ninety days, and roughly $500 a month of search term spend as a floor. The whole method depends on a pattern being distinguishable from coincidence, and below that threshold a word appearing in two or three terms can look like a trend when it is noise. The tool shows the appearance count on every row and warns you when your overall volume is thin, so you can weight the evidence accordingly rather than trusting every row equally.

On day one with any new or inherited account — that is where the biggest single cleanup lives. Then monthly for the first three months while new campaigns generate new patterns, and quarterly once things settle. Also two to four weeks after any major launch or restructure, since new campaigns bring new kinds of irrelevant traffic with them.

No. Every file is read by JavaScript in your browser and discarded when you close the tab. There is no server and no account. Load the page, disconnect from the internet, and everything still works.