Search term analysis with AI: pre-filter the export or send every row?

Pre-filtering a search term report before handing it to an assistant bakes your assumptions into the input, so the AI mostly ends up confirming what you already believed. A lot of people cut by click count or cost so the model only sees what they consider meaningful, but those cuts are exactly where the surprises hide. The odd terms it groups together are often ones you'd have dropped by hand.

Sending the full export is slower and eats context, I'll give you that. And raw data can bury the signal if the file is huge, so you may need to split it by ad group or date range instead of trimming rows. Where do you draw the line?

Nova, I'd push back on the click cutoff more than the row count. In e-commerce, filtering by clicks is where the damage happens, because conversion value per click swings so much with margin. A term with 3 clicks and one sale of a 180 dollar item is worth keeping, while a term with 60 clicks and zero sales is pure waste. A click threshold would flag the first and keep the second.

So I'd only drop rows with no spend and no conversion value, then sort the rest by cost. For a Shopping or Performance Max account, I'd split by campaign before anything else. PMax search term data mixes brand, category and product-type queries, and if you lump a branded campaign in with a generic one, the generic terms look better or worse than they really are. Splitting first keeps the model comparing like with like, and the file stays manageable without you deciding what matters.

Priya, splitting by campaign makes sense, but I'd be careful with the "no spend and no conversion value" cut on recent data. If a term picked up clicks in the last few days, zero conversions may just mean the sale hasn't closed yet, especially with longer consideration cycles. Pull the report for a window that gives conversions time to land, or at least treat the most recent week as provisional. Otherwise the model will confidently label a fresh, good term as waste.

Agreed on the recency lag, and conversion delay makes it worse than most people expect. With a 30 day click-through window, a term showing zero sales from last week can still look fine once the lagging conversions come in. So filter on a date range where even the oldest rows have had that full window to mature.

Set the maturity window per conversion action, not once for the whole account. A purchase might close within a day or two, while a quote request or lead form can take a few weeks, so a single 30 day rule can fit the first and still be too short for the second. The conversion lag report in Google Ads shows how long each action typically takes, which gives you something concrete to base the window on. Do you usually mix purchase and lead actions in the same export, or keep them apart?

I'm not sure the conversion lag report should set the window for lead actions, at least in local services. It only shows how long Google takes to record an action, but for a plumber or clinic the real delay usually comes after that. A booked call or form fill counts on day zero, then someone qualifies it and the job closes a week later. So the window that matters is the one where a lead turns into revenue. If you aren't importing offline outcomes, treat the lag report as the minimum, and don't trust any lead term that looks dead inside 30 days.

Marco, I don't think 30 days works as the cutoff for calling a lead term dead. All a window tells you is that a term is old enough to judge, not whether it's any good. A term with 20 form fills and zero qualified leads after six weeks is dead, and a term with two fills that both turn into jobs isn't, whatever the day count says.

So I'd use the 30 days only to decide when you're allowed to look at a term. The actual call should come from the qualification stage, which is why the offline outcome import matters more than the window.