Google Search now has two generated layers, and they are easy to blur. AI Overviews, which Google began rolling out to everyone in the US in May 2024, put a generated summary above the usual results. AI Mode, which Google announced as a Labs experiment in March 2025 and opened to everyone in the US in May 2025, is a separate conversational mode: you ask a question, get a written answer with links to its sources, and ask follow-ups instead of scrolling a page of results.
For local lead generation, both are a layer between a person with a problem and a business that solves it, and the businesses named in that layer are chosen by a different process from the one that decides rankings. AI Mode goes furthest, because the answer is the page.
This post is about what actually changes for lead gen, and what does not. It avoids predicting where the interfaces go next, because that changes faster than any advice about it survives. Google changes both products often, so check the dates on the announcements linked here before you repeat a detail to a client.
How AI Mode presents local businesses
Google's announcement of AI Mode describes a "query fan-out" technique: instead of one search, it runs several related searches at once across subtopics and data sources, then writes one answer from what comes back. Google also says AI Mode draws on real-time sources such as its Knowledge Graph and information about the real world, not only web pages.
What that means for a local question is a written recommendation rather than a ranked list. Run your own category and city in AI Mode and note three things: which businesses are named, which sources are linked, and whether business details such as a rating, opening hours or a phone number appear beside the names. Where they do, that is Google's Business Profile data at work, which is why the profile fields that feed AI answers matter here as much as anywhere.
The follow-up box matters for lead gen too. A person who asks "which of these can come out tonight?" narrows the shortlist again without ever seeing a results page. The businesses that state their hours and response times plainly have something to survive the second question with.
What AI Mode means for calls and attribution
Google's guidance for site owners on AI features says sites that appear in AI features such as AI Overviews and AI Mode are included in the overall search traffic Search Console reports. Check that page for the current reporting options before you promise a client an AI Mode number.
A visit is the easy case. A call placed straight from the answer, or a person who reads a name and searches for it later, may never reach your site at all. The attribution section below plans for that.
The structural change
Classic local search gave a person a page of options: a map pack with three businesses, then ten organic results, then more. The person did the choosing.
A generated answer does the choosing first and presents the result as a recommendation. A few names, a sentence each, and usually a suggestion about what to do next.
Two consequences follow, and they are the only two that matter.
The shortlist got shorter. There is no equivalent of position 8. A business is in the paragraph or it is not. The AEO vs SEO post covers what a shorter shortlist does to the work.
The click got optional. A person can read a recommendation and act on it without visiting anyone's website. They call the number, or they search the business name directly, or they open the map. From your analytics, the visit that started the whole thing is invisible.
What AI Mode does to lead gen economics
Lead volume gets lumpier
When ten businesses shared a page of results, demand spread. When a few names share a paragraph, demand concentrates. The businesses in the paragraph get more than their old share; everyone else gets a step change downward.
This makes lead gen as a business both better and riskier. Being one of the named businesses in a category and a city is worth more than a first-page ranking used to be. Not being named is worth close to nothing.
Attribution gets worse before it gets better
The lead arrives as a phone call. There is no referrer, no UTM, no landing page in the path. Ask the caller how they found you and expect to hear "I found you on Google", because from their point of view they did.
You cannot fix this with tooling. You can partially close it:
- Add one question to the intake script and log the answer. It is crude and it is the best signal available.
- Watch branded search volume. A rise in people searching your business name by itself, with no campaign running, is the shape a recommendation leaves behind, though it does not tell you which surface produced it.
- Watch direct traffic and calls together rather than separately.
- Track your visibility score independently of leads, and correlate the two over months rather than weeks.
Say all of this to a client before you start, not after. It is a reasonable limitation and an indefensible surprise.
The audit becomes the product
Because attribution is weak, the thing you can actually show a client is the measurement: for these questions, on this day, you were named this often and these competitors were named this often.
That measurement has to be done carefully to be worth anything. Same questions every time, logged out, fresh conversations, several phrasings rather than one, and the raw answers kept so anyone can check them. Our methodology page sets out how we do it, including the parts where the method is approximate, because a measurement you cannot audit is a number you should not bill for.
What still works
A useful amount of the old playbook survives intact, because generated answers are built from ordinary web sources. Google's guidance for site owners says there are no extra technical requirements to appear in AI Overviews or AI Mode: a page that is indexed and eligible to show in Search with a snippet is eligible to be linked.
Google Business Profile completeness. Still the first hour to spend. Category, services, hours, description. It feeds the surfaces directly and it feeds the pages those surfaces read.
Reviews, with a twist. Volume and rating still count. What is newly important is what the reviews say, because generated answers quote themes. Reviews that repeatedly mention a specific service or a specific speed give a model a clause. Generically positive reviews do not.
Citations and consistency. Same name, address and phone everywhere, for the same old reason: it is how any system decides two records are one business.
Pages that answer a question plainly. A page that states what you do, where, for whom, and roughly what it costs is easy to quote. A page of adjectives is not.
Local links. Still the mechanism by which you enter the small set of pages that get read when an answer is composed.
None of that is new. The change is which of it matters most: being on a handful of the right pages now matters more than ranking on a page of ten.
What changes in the work
Three practical shifts.
Optimise for the question, not the keyword
Keyword research gives you "emergency plumber city". Question research gives you "who can come out tonight to fix a burst pipe". The second is what people type into an assistant, and it is what a page should answer in its first paragraph.
Write pages that answer whole questions in short declarative sentences. That is not a style preference: short factual sentences survive being compressed into a recommendation, and paragraphs of qualified marketing prose do not.
Chase the citation set, not the whole web
Find the pages the answers actually cite for your key questions. There will not be many. Getting onto those specific pages is work with a visible target. A general link building programme is not, which is the whole difference. ChatGPT cites its sources in a similar way, and how to show up in ChatGPT recommendations covers the checks that are specific to it.
Publish numbers
Response times, price ranges, guarantee terms, years trading, service radius in miles. Anything with a number in it is quotable. "Competitive pricing" is not. This is uncomfortable advice for businesses who prefer "call for a quote". It is also the cheapest thing on this list, because the numbers already exist inside the business and publishing them costs nothing but the decision.
What this means if you sell leads rather than services
Lead sellers should be paying closer attention than agencies, because the concentration cuts both ways.
If you own the businesses being named for a category in a city, your cost per lead falls and your defensibility rises. If you own a network of thin sites that ranked on volume, the generated layer sits above all of them and answers the question without a click.
The practical move is to stop measuring your position and start measuring your mentions. Take the eight questions a buyer asks in your category, run them monthly, and record which businesses get named. That list is the market. If your properties are not on it, your traffic is being intercepted before it reaches them, and no amount of on-page work on a page nobody opens will change that.
A cheap monthly routine
About twenty minutes by our own estimate, once a month, per market you care about.
- Eight questions, written down once and never changed.
- Fresh conversation, logged out, at least two assistants, with AI Mode as one of them if your customers search on Google.
- Record every business name in every answer.
- Score: mentions divided by answers, per business.
- Note any new name that appeared and go look at what changed on their profile.
Step 5 is the easiest one to skip and it is where the learning is. When a competitor appears from nowhere, go looking for one identifiable change: a new directory listing, a corrected category, or a roundup post that named them. The 15-minute manual audit is the long form of this routine, with the scoring spelled out.
Doing that for one market is twenty minutes. Doing it for forty cities and five trades is the thing that needs automating, which is what AIgnored is built for, with two differences. A scan asks its own fixed set of 8 broad recommendation questions per model, listed in every scan's why drawer, rather than the eight you wrote down. And it asks the models through their APIs, so AI Mode itself is not one of the surfaces it measures. It scores every business in the city and flags the businesses with real reviews that never get named as the ones worth contacting. It runs on your own provider keys, so a scan costs what the provider charges. The pricing page includes a free plan for a single city.
The summary
AI Mode and AI Overviews shorten the shortlist and hide the click. The shortlist is chosen from a small set of ordinary web pages and Google's own business data, which means the work that gets you named is mostly work local SEO already knew about, aimed more precisely.
What genuinely breaks is attribution. Plan for that in how you sell and in how you report, and the rest of it is a smaller, more tractable version of a problem the industry has been working on for twenty years.