What you'll learn in this article
- What actually happens to impression share when a keyword bid sits below the auction floor
- Why an underbid keyword looks healthy in every column except one
- The diagnostic sequence I run before touching a single bid
- How the same silent error shows up inside automated bidding as a low target
- What I infer from a keyword with strong CTR and near-zero impressions
The most expensive mistake I find in inherited accounts isn't overspending. It's a keyword that has been bid at half the auction price for eleven months, quietly serving forty impressions a week on a term the business built its whole campaign around. Nobody flagged it, because an underbid keyword breaks nothing visible. It just stops showing up.
This is what I've learned about how to bid on keywords in google ads by working backwards from that failure what a too-low bid does to your impression share, why the account hides it so well, and the sequence I run to correct it without blowing up the cost per conversion in the process.
What a bid below the floor actually does
A keyword bid isn't a price you pay. It's an entry ticket to an auction you may or may not clear. Below a certain threshold you don't get a worse position you get no position at all, in an increasing share of the auctions you were eligible for.
The behaviour I see is not linear. A bid at 80% of the going rate might cost 30% of impression share. A bid at 50% often costs 85% or more, because at that level the keyword only clears the auctions nobody else wanted: odd hours, thin geographies, low-competition long-tail variants. Volume collapses, and what survives is a biased sample of the traffic you thought you were buying.
The column that tells the truth
Search lost impression share (rank) is where this shows up, and it is the only column in the account that does. Google's documentation on impression share data defines it as the share of time your ads didn't appear on Search because of poor Ad Rank in the auction, and notes it's enabled through the competitive metrics columns on the keywords view which is exactly where most accounts have never switched it on.
Two details from that documentation matter operationally. The metric won't display at ad group level if the campaign ran out of budget at any point in the date range, so a budget-capped campaign will hide its rank problem entirely. And impression share updates within a day or two, which means a bid change made this morning cannot be judged this afternoon.
What the underbid keyword looks like
In my experience it presents as: low impressions, respectable CTR, a conversion rate that looks fine, a cost that's well under budget, and lost IS (rank) somewhere north of 70%. Every metric except the last one says the keyword is behaving. That's the trap the sample is small and self-selected, so it looks clean precisely because it's starved.
Why the error stays silent for months
Bidding on keywords in google is one of the few operations in the platform with no failure state. There is no warning, no disapproval, no diagnostic flag. Four things keep it invisible.
1. The account rewards it on every dashboard metric
An underbid keyword has low spend and often a flattering cost per conversion on its handful of conversions. On a weekly report it reads as an efficient keyword. Nothing in the default column set says "this is efficient because it is barely running".
2. Estimates are set once and never revisited
Most bids I inherit were set at launch against a planning figure and never touched. Auction prices drift upward continuously; a bid that cleared comfortably in Q1 can be under the floor by Q4 without anyone changing anything. Comparing the current bid against a fresh keyword cost estimate is the fastest way to see how far the market has moved away from you.
3. Budget masking
When a campaign is budget-limited, lost IS (rank) becomes unreliable or unavailable at ad group level. So the campaigns most likely to have starved keywords are also the ones where the diagnostic column goes dark. I resolve the budget constraint first, wait for clean data, then look at rank.
4. Automated bidding hides it behind a target
The same error exists inside smart bidding it just wears a different name. An overly aggressive target CPA or target ROAS makes the system bid conservatively, and the keyword loses auctions exactly as a manual underbid would. The column reads the same. When I see rank loss on a campaign using an automated strategy I look at the target before anything else, since smart bidding is only permitted to spend as much as the target lets it.
How I correct an underbid keyword
The correction matters as much as the diagnosis, because raising bids carelessly turns a visibility problem into a cost problem within days.
Step 1 Separate rank loss from budget loss
I enable both lost IS columns before doing anything. If budget loss is meaningful, the rank figure isn't trustworthy and the bid isn't the variable I should be touching yet.
Step 2 Check quality before spending
Ad Rank isn't bid alone. Where relevance and landing page experience are weak, the bid required to clear the auction is inflated, and I'm paying a premium for a fixable problem. If the keyword's quality score is poor, I address that in parallel rather than buying my way past it.
Step 3 Raise in increments, not in leaps
I move bids 15-20% at a time and hold for a week. Doubling a bid tells you nothing useful: you can't attribute the resulting change to any specific level, and you've committed to the new cost before knowing what it bought. Incremental steps let me watch lost IS (rank) fall and stop at the point where it stops falling meaningfully.
Step 4 Decide what impression share you actually want
100% is almost never the right answer. The last twenty points of impression share are the most expensive to buy, because you're outbidding everyone in the auction, including advertisers with better economics than yours. I usually target the range where incremental impression share still comes at an acceptable incremental cost per conversion, then stop.
Step 5 Re-check on a schedule, not on suspicion
Auction floors move. I put lost IS (rank) into a recurring weekly optimisation routine so drift gets caught at 20% rather than discovered at 80% a year later.
What I infer from the pattern
High CTR plus low impressions is an underbid, not a great keyword. The ads people love are the ones served in the least competitive slivers of the auction. I read that combination as a bid problem until proven otherwise.
A keyword with zero data isn't unproven, it's untested. I've seen accounts pause terms for "not converting" on thirty impressions accumulated over a quarter. That keyword was never given an auction to lose.
Sudden rank loss across many keywords is a market event, not an account event. When it moves everywhere at once, a competitor changed budget or a new entrant arrived. That's a pricing decision to make, not a bug to fix.
Lost IS (rank) rising while CPC stays flat means the floor moved without you. Your cost didn't change because you're still winning the same cheap auctions you just stopped being eligible for the rest.
Under automation, the target is the bid. Every conclusion above transfers directly to smart bidding, with target CPA or ROAS as the lever. The failure mode is identical; only the control surface differs.
What I stopped doing
Judging keywords on efficiency metrics alone. Cost per conversion on a starved keyword describes an accident of which auctions you happened to clear, not a performance level you can scale.
Raising bids and lowering them again in the same week. Impression share takes a day or two to settle and the auction takes longer than that to respond. Reversing before the data lands means learning nothing twice.
Chasing 100% impression share on brand-adjacent terms. The final stretch is bought at a price the incremental conversions rarely justify.
Treating manual and automated bidding as different problems. They aren't. A conservative target starves a keyword exactly like a low max CPC does, and I diagnose both from the same column.
The practical takeaway
If you take one operational habit from this: enable Search lost IS (rank) at keyword level and look at it before you look at anything else. It's the only metric that reports on the auctions you never entered, and every other column in the account describes only the ones you won.
Bidding well isn't about finding a correct number. It's about knowing which of your keywords are competing and which are quietly sitting out, then deciding deliberately which ones deserve the price of entry.