Reviewing the search terms report for obvious waste is repetitive work, and repetitive work is what scripts are for. A Google Ads script can pull the report on a schedule, apply a rule like "block terms over a spend threshold with no conversions," and either email you a suggestion list or add the negatives itself. The free and paid community scripts that do this are everywhere. The question worth answering is not whether you can automate negation — you can — but which part of the job is actually safe to hand over, and which part will quietly cost you money if a rule makes the call instead of a person.
The short version: automate the unambiguous waste, keep the ambiguous decisions human, and start every script in a mode that only suggests until you trust it. This post is about where that line sits and how to build the guardrails that keep a scheduled script on the safe side of it.
What these scripts actually do
At their core, negative-keyword scripts all do the same three things: read the search terms report, test each term against a threshold rule, and act on the ones that qualify. Where they differ is the last step. Some are deliberately cautious — one widely shared script suggests negatives for search terms with significant clicks but little to no conversions and emails you a spreadsheet to review, adding nothing itself. Others write straight to the account: a well-known "eliminate wasted spend" script moves non-converting search queries into a negative list once they pass a spend threshold you set, scheduled to run daily.
That difference — suggest versus apply — is not a minor configuration detail. It is the entire safety boundary. A suggestion script cannot hurt you; the worst case is you ignore an email. An apply script changes your account unattended, on whatever cadence you scheduled, using only the evidence its rule could see. Both are built on the same report and the same threshold logic, so it is tempting to treat them as interchangeable and go straight to the automatic one. Resist that. The capability is cheap; the judgement about when to let it write is the whole game, and it is worth earning the automatic mode rather than enabling it on day one.
What is safe to automate
The band of waste that is safe to automate is the unambiguous head: terms that spent clearly more than they should have and returned nothing, over a window long enough that the nothing is real. If a term has burned several multiples of your target cost per acquisition with zero conversions across 60 or 90 days, that is strong evidence, and a rule can act on it as well as a human would. This is exactly the work worth automating, because it is high-volume, boring, and easy to fall behind on — the kind of task where a human reviewer gets fatigued and starts rubber-stamping, which is itself a source of error.
The key is that the threshold has to be set where crossing it is genuinely conclusive, not merely suggestive. A rule that blocks terms after a single zero-conversion click is not automation, it is vandalism on a timer — one click is almost no evidence, and it is one of the most common ways to over-prune an account. Set the spend bar high and the window long, so that by the time a term qualifies, any reasonable practitioner looking at it would reach the same verdict. If you would hesitate to block it by hand, the rule should not block it either. Automate only the decisions that are not really decisions.
The guardrails against negating converting terms
The expensive failure of an automatic script is negating a term that was quietly helping, and it happens for a specific, avoidable reason: naive rules look only at last-click conversions over a short window. Three guardrails prevent it. First, set the spend threshold well above target CPA so a term has clearly failed, not just had a quiet month. Second, use a conversion window long enough that a delayed conversion is not read as a zero — in long sales cycles a term can convert weeks after the click. Third, and most overlooked, exclude terms with assisted conversions or conversion value, not just last-click conversions. A term that assisted a sale credited elsewhere looks like pure waste to a rule that only counts direct conversions.
Even with those guardrails, some terms simply should not be handed to a rule at all. The ambiguous ones — queries that could be a buyer or a browser depending on context the script cannot see — belong in a human review queue, not an automatic block. A good automated setup is therefore two-tier: the script auto-handles the unambiguous high-spend zeros and routes everything borderline into a report for a person to judge. That split is what lets you automate safely — you are not choosing between all-manual and all-automatic, you are letting the rule take the easy half and reserving your attention for the half where attention actually pays. A script that tries to decide everything will eventually throttle good traffic, and a pure-manual process will fall behind; the tiered one does neither.
Scope, match type, and close-variant blindness
Automated scripts usually add negatives at the ad group level, and that caution is correct. Ad-group scope is the narrowest: a mistaken negative there affects one ad group, so the blast radius of a bad rule is small. The opposite extreme — an account-level shared list — is powerful precisely because one entry blocks a term everywhere, which is the last thing you want an unattended rule touching. Let the script work at ad group or campaign level, and keep account-wide exclusions a human decision. The convenience of automating the broadest scope is not worth a rule silently cutting a term across your whole account on a bad threshold.
There is also a limit no script escapes: negatives block strings, not meanings, and they do not match close variants. A script that adds "cheap" as a negative does nothing about "cheapest" or the misspelling "chaep" unless those exact strings also qualify under the rule and get added separately. So even a well-run automatic script leaves the plural, synonym, and typo coverage incomplete, which is one more reason the human review tier matters — a person expanding a blocked term into its variants is doing something the string-matching rule structurally cannot. Automation narrows the gap; it does not close it.
Rolling it out without regret
Start in suggestion-only mode, always. Run the script so it emails what it would have blocked, and read that output for a few weeks against your own view of the account. You are checking two things: that it is catching real waste you would have blocked anyway, and — more importantly — that it is not proposing anything you would have kept. Only when a run goes by with no suggestion you disagree with should you consider letting it write to the account, and even then, log every change. A script that records each blocked term to a sheet gives you the audit trail to catch a bad rule before it compounds, and to reverse it if a negated term turns out to have mattered.
Treat the script as one layer in a wider hygiene routine, not a replacement for it. It automates the repetitive bottom of the funnel so your attention goes to the parts that need judgement — the ambiguous terms, the new product lines, the pattern-level fixes an n-gram analysis surfaces, and the periodic read of whether your wasted-spend numbers are actually falling. Automation earns its keep when it removes drudgery and leaves the decisions with you. It stops earning its keep the moment you let it make the decisions too — and the whole point of the suggestion-first rollout is to make sure you never cross that line by accident.