Skip to main content

Understand its recommendations

Hero AI's responses read like polished prose. Underneath, data lookups and reasoning steps build them. An answer mixes measured facts and the model's interpretation. Know which is which. Then you know what to act on directly, and what to verify first. The short version: trust the specific numbers, test the causal claims.

Where the answer comes from​

Hero AI builds every answer from these sources:

  1. Live Google Ads data. Campaigns, ad groups, keywords, search terms, ad copy, settings, and performance for your active account, read when you ask. This is the most reliable layer. Numbers that Hero AI cites come from here, or from what you typed or attached.
  2. Your product. The name and website of your product. When Hero AI writes ad copy or a landing page wireframe, it also uses your full product context, with its proof and offers. Hero Marketer builds product context from your website. It includes any edits that you made in Settings > Products. To open Settings, click your profile picture at the bottom of the sidebar, then click Settings.
  3. What you give it. Your messages, the files that you attach, and the web pages that you link in the chat.
  4. General knowledge of Google Ads and B2B SaaS. Built into the model. Useful for framing, but not specific to your account.

When Hero AI explains why something happens, the model reasons over the live data and its general knowledge to build that explanation. The explanation is not a measured fact.

Some reports carry extra interpretation rules:

  • Landing page reports group tracking URL variants into real destination pages first. Keyword rows in that report come from URL tracking parameters when present. See Read landing page reports.
  • Simulation results are Google's historical simulation points for a bid, budget, Target CPA, or Target ROAS change. Treat them as planning evidence, not a promise of future results.
  • Click diagnostics are narrow checks of one day within the last 90 days. They are useful for lead quality, GCLID, or traffic quality questions, but too noisy for a normal campaign audit.

What is a measured fact versus an inference​

Every recommendation mixes two kinds of statement, and they earn different levels of trust:

  • Observations are facts pulled from your account. "Your CPA on the alternative to jira campaign is $245 over the last 14 days" is an observation. You can confirm it in the dashboard. Trust these.
  • Inferences are conclusions Hero AI draws from the observations. "The high CPA is likely driven by a few high cost keywords that aren't converting" is an inference. It might be right. It might be wrong. Verify these before acting.
  • Simulation points sit between the two: they come from Google Ads, but they describe modeled tradeoffs from historical data. Act on them only when observed performance points the same way.

How Hero AI weights what it finds​

Hero AI does not treat every blip as equally meaningful. It checks whether a finding holds across three time windows. By default, these are the last 30 days, the 30 days before, and a longer history before that. Then it weights its recommendations toward the findings that hold up:

  • A finding consistent across all three windows is usually structural and worth acting on.
  • A finding that shows up in a single window only may revert on its own. Hero AI flags it as worth watching. It says it is too early to act.
  • A setting problem, for example broken tracking or Search Partners turned on, is flagged at once, whatever the windows show.

The evidence that Hero AI needs grows with the size of the action. An optimization needs at least 2 windows that agree. A budget increase or an expansion needs all 3.

The supporting evidence section under each recommendation shows the per window figures, so you can confirm which case applies.

When a recent change in your account lines up in time with a metric shift, Hero AI scores that change as a primary suspect, a secondary factor, or ruled out. This shows you exactly what it attributes the move to. Then you can revert the right thing and not pause the affected campaign.

Hero AI reads ad strength as Google's checklist of the number and variety of your headlines and descriptions, not as part of the auction. It treats Average or Good ad strength as fine, and it does not tell you to unpin lines only to raise the score. See How Hero AI reads ad strength.

How to verify before acting​

Three quick checks before you act on a recommendation:

  • Cross check the dashboard. Hero AI's numbers should match what the dashboard shows for the same date range. A mismatch is usually a date range or filter difference, not a data error. Hero AI uses the last 30 days by default. The dashboard uses the last 7 days by default.
  • Open Google Ads. For recommendations about individual keywords, the search terms report and the performance view for each keyword are the source of truth. Hero AI summarizes. Google Ads shows the underlying rows.
  • Ask Hero AI to show its work. "What specific keywords are you talking about? Show me the numbers." It lists the keywords and metrics behind the recommendation.

When recommendations are reliable, and when they are not​

Hero AI's recommendations are more reliable when:

  • A pattern holds in all three windows.
  • You state a target, so Hero AI measures against it.
  • Conversion tracking works.

They are less reliable when:

  • The campaign has only a few days of history.
  • The pattern shows in one window only. Hero AI then says it is too early to act.
  • Conversions are zero, so Hero AI can judge only CTR and CPC.

Low conversion volume is normal for B2B SaaS. Hero AI does not wait for a fixed number of clicks or conversions. It checks whether a pattern holds across the windows.

When Hero AI is wrong​

Hero AI is wrong sometimes. The common failure modes:

  • It misses context outside Google Ads. A CPA spike might come from an industry shift or a competitor's launch that Hero AI cannot see.
  • It can miss a very recent change. A change that started only in the last few days is small inside a 30-day window.
  • It can be confidently wrong. Polished prose makes a guess sound like a finding. Treat specific numbers as facts. Treat causal claims as hypotheses to test.

When Hero AI is wrong, push back: "Are you sure that's what happened? What's the underlying data?" That usually surfaces the gap, or makes Hero AI correct itself.

What to do with a recommendation​

Most recommendations come with one of three paths:

  • Hero AI runs it for you. Hero AI can pause a campaign, add a negative keyword, change a budget, or apply new ad copy. It stages the change as an approval card in the chat, and runs it when you click Approve and apply. See Actions Hero AI can take.
  • Hero AI drafts a wireframe. For a landing page issue, Hero AI drafts a wireframe for your designer. The wireframe changes nothing until you build the page. See Landing page redesign.
  • You run it in Google Ads. For changes Hero AI cannot make, for example when you restructure targeting or edit conversion tracking, the recommendation gives you detailed instructions to follow yourself.

Two rules of thumb regardless of path:

  • For changes you are unsure about. Test on one campaign first, give it a week, then evaluate.
  • For big strategic moves. For example, if you kill a campaign, restructure targeting, or double a budget, get a second opinion. Ask Hero AI to argue the opposite ("What's the case for not doing this?") or run it past your team before you approve the change.

Next​