A lead gen account will optimise toward whatever you tell it a conversion is. Tell it a conversion is a form fill, and it will find you the cheapest form fills on the internet — students, bots, competitors, the permanently curious. The escape is a quality feedback loop run every single week, and the reason most accounts never run one is that it is tedious. That tedium is exactly what an agent connected to your live account removes — it works the loop at the speed of a sentence, week after week, and never tires of the grind.
The lead gen trap: cheap forms, expensive junk
The trap has a respectable face: cost per lead falling quarter over quarter. Meanwhile sales works ever harder to find a real buyer in the pile, and the true number — cost per qualified opportunity — climbs. Nobody sees it because the account measures fills, the CRM measures deals, and the two systems never talk. The account is not broken; it is obedient. It needs a better definition of success, and someone with time to enforce it weekly.
Decide which conversions are worth paying for
Every fix starts in the conversion actions screen. The working hierarchy: qualified outcomes — a booked call, a verified phone lead, an application completed — set as primary, so bidding optimises toward them. Raw form fills demoted to secondary: observed, reported, never optimised for. If your volumes are thin, a middle tier helps — fills that passed validation, imported as a distinct action. Ask the agent to read your current conversion actions and tell you what bidding is actually optimising toward; the answer surprises most advertisers. Re-ranking the actions you already have happens in the Google Ads conversion settings — the connector never edits, demotes or removes an existing conversion action — with the agent supplying the plan and reading the result, and it is the highest-leverage hour you will spend on the account.
If the qualified outcome is not counted at all yet, the connector builds the tracking rather than describing it. As of v2.16.0 it runs the whole chain, every write proposed first and applied only on your yes: a conversion setup audit that says whether today's tracking can be trusted; a new Google Ads conversion action for the outcome, primary or secondary as you choose; the conversion tag and the trigger that fires it, staged in Tag Manager on a site that already loads it; and, where the outcome is an Analytics event, marking it as a key event and linking the Analytics property to Google Ads. Two steps stay with a person: publishing the Tag Manager container, and importing an Analytics key event as a conversion in Google Ads.
The quality feedback loop: CRM outcomes back into bidding
The loop is simple to describe: every week, what the CRM learned about last month's leads flows back into the account. Leads that became opportunities push their campaigns up; sources of junk get starved. Enhanced conversions and offline import automate the data path once set up through Google's native routes. The agent's role is the enforcement: each week it reads performance by qualified conversions, names the campaigns and terms producing junk, and drafts the responses — negatives, budget shifts away from junk sources, a paused rebuild of the ad group that attracts the wrong intent. You approve; the loop actually closes. Accounts run this loop for two weeks easily. The agent makes week forty look like week two.
Search terms discipline for lead gen intent
Lead gen queries betray intent in their grammar. "Free", "salary", "jobs", "course", "template", "example", "DIY" — each has a meaning, and most of the meanings are not buyers. Each week the agent hunts the spend that buys nothing — sorting by conversion-free cost and clustering the offenders: job seekers, students, researchers, people looking for the thing you sell but free. It drafts the negative keyword additions with the evidence attached — this term, this spend, zero qualified leads — and you approve the batch in one click. Discipline also means restraint: terms that convert oddly but cheaply get a note, not a negative, until the CRM verdict arrives.
Spam and bot-fill patterns an AI catches early
Junk has signatures. A campaign whose conversion rate doubles overnight with no change you approved. Form fills clustering at 3 a.m. from one region. A display placement suddenly producing conversions at one-tenth the usual cost — always a red flag, never a miracle. Gibberish-name leads arriving seconds after click. Reading for these patterns daily is unreasonable for a human and trivial for an agent asked weekly: what changed abnormally, where, and does the pattern look like intent or automation? Early detection matters because bidding learns from junk conversions — every week a bot pattern runs, Smart Bidding gets more confident that bots are your customer.
The weekly lead-quality review, as prompts
One conversation, forty minutes. Which campaigns spent the most per primary conversion this week — and how does that compare against our target CPA? Which search terms spent with zero qualified outcomes — draft the negatives. What does performance look like measured only by imported qualified actions? Any conversion anomalies that fit spam patterns? Which conversion action went quiet — tracking breaks silently. End with the drafted changes queued for approval and a two-paragraph summary for whoever asks how ads are going.
Where the approval line sits
Reads and the weekly review change nothing — allow them once (in Claude, set the read-only tools to Always allow) and they stop prompting. Negatives and paused builds are fast yeses — they cannot spend. The firm line: budget moves, pausing anything live, and enabling anything — each waits for the click of whoever owns the pipeline number. A bidding strategy switch sits on that line too: the connector can make one, and because it resets learning it waits for the same click. So does every step of conversion setup. Editing or demoting a conversion action you already have is not in the toolset at all, and stays deliberate in the Google Ads UI. And the floor under everything: the connector cannot delete. The remove tools are never exposed, and a REMOVED mutation is refused server-side, so the worst any approved action can do is a change the account itself can put back — a pause, a budget, a bid, a keyword — never a deletion.
The boundary: your CRM, your form, your routing
It does not see your CRM. The quality verdicts — which leads were real — enter either through Google's import pipeline or through you, in conversation. It cannot fix your form, your landing page or your lead-routing rules, and it cannot make sales dispose leads faster. It reads Google Ads live — with Analytics, Search Console and Tag Manager when you connect them — and nothing beyond Google's own products. And it will not invent a quality signal you refuse to define: if nobody decides what a qualified lead is, the loop has nothing to feed back — that decision stays yours.
A pilot plan for a lead gen account
Day one: connect and run the read-only audit — pay attention to the conversion action findings first, spend waste second. Day two: define the quality hierarchy and apply it in the conversion settings, with the agent drafting the plan, proposing the conversion action and staged tag for any outcome not yet counted, and confirming what bidding now optimises toward. Days three to five: the first search-term sweep with negatives, the spam-pattern read, and a budget proposal that moves money toward campaigns producing qualified outcomes. The trial is 7 days on the full Pro plan for a new workspace, no card, then it drops to the Free plan (one account) — not a lockout. If your leads feed a long sales cycle, the B2B use-case extends this loop to pipeline data. Start free and run the first quality review this week.