What is Negative Phrase?

Negative phrase stops your ad serving on any query that contains a given sequence of words, in that order, whatever surrounds it. One entry covers the queries you have already seen and the ones you have not seen yet, which is why it is the leverage tool in exclusion work.

Exact against phrase

Why themes, not terms

Wasted spend is usually not concentrated in a few expensive queries. It tends to spread across a long tail of cheap ones that share a single wrong word. Term-by-term review rarely finds it, because no individual term spends enough to attract attention. Grouping the search term report by word groups, single words then pairs then triples, and totalling spend and orders per group, exposes the shared word and lets one negative phrase retire the whole theme.

Drill down before you commit

A negative phrase is close to permanent in its effect, because you stop seeing the traffic it removes and therefore stop being able to evaluate the decision. Two safeguards:

• Exclude a word group only when the large majority of the terms beneath it are genuinely irrelevant. If the report shows a mix, the group is ambiguous.
• When a single word is ambiguous, exclude at the pair level instead, and scope it to the campaign or ad group where it is wrong rather than to the account.

Worked example

Suppose you sell adult cycling gloves and the search term report shows the word kids appearing across many low-spend queries, none of which produced an order. Every term beneath it is irrelevant, so kids is a sound negative phrase.

Now suppose the word is gift. Some gift queries convert and some do not, so the single word is ambiguous. The better move is a pair such as gift wrap, which is reliably wrong, rather than blocking every gift shopper.

Cadence

The first pass over a newly managed account usually returns the largest haul. After that, monthly for the first quarter and quarterly at a steady state is normally enough, with an extra pass two to four weeks after a launch or a large bid change, because both alter the query mix. Run the mirror image at the same time: word groups with strong order counts and low ACoS are harvest candidates and listing copy candidates.