A keyword in a search campaign is not a filter that only admits identical queries. It is an instruction that the platform interprets, sometimes generously.
Match types describe intent, not exact strings
Broad settings allow a keyword to trigger against queries the platform considers related in meaning, including synonyms, related concepts and reformulations.
Even the strictest setting admits close variants such as plurals, misspellings and reordered words, which is a deliberate design choice rather than a defect.
The rationale is that a large share of searches are new phrasings never seen before, and rigid matching would leave most demand unreachable.
The search term report is the real inventory list
What an advertiser buys is visible only in the report of actual queries served, which frequently bears limited resemblance to the keyword list.
Reviewing that report is where irrelevant traffic is identified, and it is the main routine task that separates a maintained account from a neglected one.
Accounts left unreviewed drift, because the interpretation widens over time as the platform learns which adjacent queries produce clicks.
Negative keywords are the counterweight
Excluding terms is how an advertiser narrows what the interpretation may reach. Without exclusions, broad settings have no boundary except relevance as the platform judges it.
Common exclusions cover research phrasing, job seekers, free alternatives and competitor names, all of which generate clicks with no purchase behind them.
Over-exclusion carries its own cost, since a blocked phrase may also appear inside a query with genuine intent that will now never be served.
Quality scoring interacts with matching
Relevance between the query, the advertisement and the destination page affects both ranking and price. Loose matching weakens that relevance by design.
An advertisement written for one phrasing and served against a distant relative of it scores worse and costs more per click than a tighter pairing.
Grouping keywords so that each advertisement addresses a coherent set of queries is therefore a pricing decision as much as an organisational one.
Automated bidding changes what matching costs
Where bids are set automatically against conversion goals, the system can tolerate looser matching because it prices each query individually rather than at the keyword level.
That only holds where enough conversion data exists for the estimates to be reliable. Thin accounts get loose matching without the pricing intelligence to manage it.
The sequence that works is to constrain matching while data accumulates, then widen it once the bidding system has enough history to judge each query on its own merits.