What is a good cost per lead from AI search?

What counts as a lead?

A lead is a person or business that took an action you agreed to count. A form sent, a call answered and a sales-ready inquiry are three different events. Pick one. Then use that same event for every channel you compare.

Google Analytics calls an action that matters to your business a key event. Google also says key events can feed Google Ads conversions, and Google Ads conversions do not show up in the standard Google Analytics reports. So two tools can count the same lead in two different ways. Choose one place to count and define the lead there.

Say you run a real estate team. A form from outside your service area and a call about a listing in your market would not both pass your sales test.

I would write the test in plain words before looking at any result. It might name the service, the market and a way to reach the person. Ask sales to apply it to each inquiry. Keep all inquiries and passed inquiries as separate counts, and never mix them in one ratio. Mixing them is the quickest way to make one channel look better than it is.

Add a field for what the person says brought them to you. People can remember an AI answer that your analytics never saw. Keep that reply separate from the recorded source. You should finish with two counts per channel: all inquiries, counted as Google Analytics key events, and passed inquiries, counted in your sales records.

What goes into the cost and the count?

Cost per lead is a plain division. Take the total cost you assign to a channel. Divide it by the number of leads assigned to that channel under your lead definition. Google Ads uses the same shape for its cost per conversion number. It divides total cost by the number in the Conversions column, and it counts only clicks that can be tracked for conversions.

The top of the fraction is where teams disagree. For paid search, the ad spend is the obvious cost. For AI search work, you may also count content work, outside fees and staff time. Decide the list once and write it down. Then apply that list to every channel. If staff time is in one channel and left out of another, the comparison is unfair.

Some work helps more than one channel. A page that wins a search click can also be quoted in an AI answer. You can give each channel a stated share of that cost. Or you can show the full cost beside both views and never add them together. Either choice works. Counting the same cost twice does not.

Dates matter too. Work you pay for this month may bring inquiries later. If you count all of the early cost and only the first leads, the channel looks poor. Repeat the same view in later periods, so you see a trend and not one snapshot.

When no leads are assigned, do not report a free lead and do not divide by zero. When the count is small, one added lead can change the ratio a lot. Show the count next to the ratio so a reader can judge how much weight it deserves.

How do you find the leads that came from AI search?

When a client asks whether AI sent any leads, I would first look at the recorded sources and what buyers say brought them in. Attribution is the rule that gives credit for a lead to a source. A person can read an AI answer, search your name later and then call. The last recorded visit shows the search. It does not show the answer that came first. That is why AI search leads are hard to count.

Google Analytics has a default channel called AI Assistant. It holds visits from sources like ChatGPT, Gemini, Deepseek, Copilot and Grok. Google's AI Overviews are the AI-written summaries at the top of some Google results. AI Mode is Google's chat-style search. Neither is in that channel. They fall under Organic Search, and Google does not split them out by feature. So the AI Assistant channel is a floor, not a full count. Organic Search holds some AI leads that you cannot pull apart.

Search Console has a similar limit. Google says traffic from AI Overviews and AI Mode is counted inside the Web search type in the Performance report. It is not a separate AI line. That report shows traffic, and a lead needs your own records.

OpenAI says ChatGPT adds the tag utm_source=chatgpt.com to referral URLs from ChatGPT search results. Look for that tag in Google Analytics. Not every ChatGPT visit carries it. A copied link or a stripped referrer, the source address sent with a visit, can land in Direct traffic, visits without a known source in analytics.

Which settings change the answer?

Two settings can change your channel totals with no change in buyer behavior. The attribution model is the rule that splits credit among the visits before a lead. The lookback window is how far back the system looks for those visits.

Google Analytics sets the default lookback window at 30 days for acquisition key events, such as a first visit. For all other key events the default is 90 days. You can switch the first kind to 7 days, and the second kind to 30 or 60 days.

A change to the reporting attribution model applies to past and future data. A change to the lookback window applies only going forward. So a new report and an old export may not match, even for the same leads.

Fix both settings before you compare channels. Write them on every report. Then a change in a report cannot pass for a change in the market. Do not add up numbers from two attribution views as if they were different leads. Keep the single lead record as the unit you count.

Do citations and clicks prove a lead is good?

Being named in an AI answer is not the same as getting a lead. Bing Webmaster Tools has an AI Performance report, announced in public preview on 2026-02-10. It shows how a site is cited across Microsoft Copilot, AI summaries in Bing and some partner services. It shows total citations, cited pages and the grounding queries, searches used to find supporting information, behind them.

Bing says the grounding query data is a sample of overall citation activity. The announcement does not describe click or lead data. Treat citations as a sign of visibility. Keep them beside your lead count and never inside it.

Google makes a related point in its own words. It says it has seen higher quality clicks from results pages with AI Overviews, meaning people spend more time on the site. That is Google's observation. It is not a lead cost, and it gives no number. Google suggests tracking conversions and time on site in Google Analytics, not only in Search Console.

If your AI search leads are fewer, check whether they pass your sales test more often. A higher cost per passed lead than another channel can still be good if the value of that lead covers the cost. Use the same lead definition, cost rules and dates for a fair comparison.

To find that value, multiply the profit on one won customer by the share of passed leads that become sales. That is the most you can pay per passed lead to break even. A cost at or below that threshold is covered by the value of the lead.

How do you judge a lead cost someone else quotes you?

Judge a quoted lead cost by what it counts, which dates it covers and which cost rules were used.

What counts as a lead? Which costs are in the cost? Which dates does it cover? Which attribution settings were used?

Google's guide Do you need an SEO suggests asking how success is measured, what results to expect and in what timeframe. It also says no one can guarantee a number one ranking on Google. A vague answer makes the result hard to assess.

The FTC Endorsement Guides say an ad that shows one customer’s result on a key point is likely read as typical. Without proof that it is typical, the ad should clearly say what results people generally get. The rule is 16 CFR 255.2(b), in the eCFR version up to date as of 2026-11-10. Ask what results people generally get, not just what one customer achieved.

For help putting those answers into an agreement, read What results should an AEO contract measure? in the related guides.

In what order should you do this?

First, choose the lead definition and write it down. Second, choose the cost list and the dates. Third, fix the attribution model and the lookback window. Fourth, set up tracking for the AI Assistant channel, the ChatGPT tag and a field for what the buyer says. Fifth, count for the period you chose. Sixth, compare with your other channels on the same basis.

Do not skip to the sixth step. Skipping ahead gives you a number that looks exact and is not. Steps one through four come before any budget decision.

How long should you wait before you decide? Before step five, pick both a minimum lead count and a number of weeks. Write both in your plan, then compare only when you have reached both. Until then, keep collecting records.

Your situation tells you where to start. If you have no tracking yet, begin at step one. If you have tracking but few leads, you are at step five, so keep counting. If you have a quote in hand, use the four questions above before you compare anything. Step one takes one written sentence. Do it before the next report.

Which guide should you read next?

Agency type: Should you hire a content agency or an AEO specialist? helps you decide which kind of provider fits your team and who should own the work.

Contract metrics: What results should an AEO contract measure? helps you choose the lead definition and other results to write into your agreement.

First 90 days: What happens in the first 90 days with an AEO agency? explains what a healthy start can look like after you hire.

Start with the agency type guide if you are choosing a provider, the contract metrics guide if you are setting terms, or the first 90 days guide if you have hired one.

What should your lead cost comparison include?

• One written lead definition for every channel
• A cost list that names shared costs once
• The same start and end dates for cost and leads
• A named attribution model and lookback window
• A lead count beside every ratio
• A separate count for leads that pass your sales test
• A field for what the buyer says brought them to you
• A plain note on what your tracking cannot see