AI changed three things in local search that you can actually verify, and left most of the job exactly where it was.
AI Overviews, the generated summaries at the top of a results page, have been caught citing self promotional listicles rather than the businesses those listicles rank. Where an AI Overview shows up on a local query, practitioners report visibility on that query falling by half or more.
And Google now uses AI to write parts of the business panel, including the services list and the description. That is why services have turned up on profiles the owner never entered.
The response is not a new discipline and it is not a product you buy. It is three moves.
Keep your Google Business Profile impeccable, because the AI features are reading it and then writing on top of it. Earn a place in the listicles the AI quotes, because that is what it quotes.
And keep measuring the Map Pack, the block of three businesses above the map, because a customer who taps Call or Directions in Maps never passed through a generated answer. This article uses a pizza restaurant throughout, since pizza is where every one of these effects is easy to see.
What AI actually changed, and what it did not
Start by separating two things that get mashed together. One is AI in the results page: generated summaries, generated profile fields, assistants answering instead of listing. The other is AI in your workflow: tools that draft your posts, your replies, your location pages.
They are unrelated problems. The first is something happening to you. The second is something you choose.
What did not change is the machinery underneath. Proximity still decides roughly 55 percent of who is eligible for the pack, before anything else is scored. No tool, audit, or AI feature moves your restaurant closer to the person searching.
After that, the profile carries the heaviest share of whatever is left, reviews come next, page content sits third, and a run of lighter signals follows. These are approximations of a system nobody has the source code for. The ordering is stable, and AI has not rewritten it.
The full taxonomy, with the numbers attached, sits in local SEO ranking factors. That also covers why the plain organic results underneath the pack lean far harder on your website than the pack ever does.
So the honest framing is this. AI has changed the shape of the page and the wording of your profile. It has not changed what makes a pizza place eligible to show up.
Change one: AI Overviews quote listicles, not pizzerias
This is the finding that should worry you most, and it is the one most AI local SEO advice skips. On local queries where an AI Overview appears, it has cited self promotional listicles rather than the business websites themselves.
Think what that means in practice. Somebody searches for the best pizza in your city. The generated answer at the top summarizes a roundup post, and that roundup was written by a directory, a food blog, or a marketing site with its own reasons for the ordering.
Your beautiful website, your menu page, your years of practice at the oven: none of that got quoted. The listicle did. And you are either in the listicle or you are not.
This flips the usual advice. Most local SEO effort points inward, at your own property. Getting mentioned in an AI Overview points outward, at other people’s pages.
It is closer to digital PR than to on page work, and it is slower and less controllable than anything else in this article. It is also not fully in your hands, which is worth saying plainly rather than selling you a package.
Change two: when an overview appears, the visibility thins out
Where AI Overviews appear on local results, practitioners report visibility drops of 50 percent or more on the affected queries. Read that carefully, because the caveats matter more than the number.
It is on the affected queries only, not across your whole keyword set. Overviews do not appear on every local search, and coverage moves constantly.
It is a reported drop from practitioners watching their own accounts, not a published Google figure, and there is no public dataset you can go check.
And it says nothing about revenue. A person who reads a summary and then opens Maps to find you has still become a customer, while showing up as a lost click.
What it does justify is a change in how you pick your battles. If a query in your set reliably triggers a generated answer, the return on ranking third organically for it just fell.
If a query reliably shows a pack and no overview, it is worth more than it was last year, not less. That is a keyword prioritization decision, and it belongs in your local keyword research rather than in a panic about AI.
There is no reliable public way to count how often you appear inside a generated answer. Anyone selling you an AI citation score is selling you an estimate. Treat those numbers as directional at best.
Change three: Google is writing part of your profile
This one gets almost no coverage and it is the one you can fix this afternoon. Google assembles the services list and the business description on the panel with AI. Entries have shown up that no owner ever typed.
Picture the consequence for a pizza restaurant. Google reads your reviews, your website and your menu, and decides you offer catering. You do not offer catering. Somebody calls at four in the afternoon asking about trays for 60 people, you say no, and they leave annoyed.
Or worse: it lists gluten free options because three reviewers mentioned a gluten free crust you tested for one summer and dropped. That is now a health conversation you did not sign up for.
The generated description has the same problem in a milder form. It is assembled from what Google has read about you, which means it inherits any stale detail sitting on an old directory page or a two year old review.
The fix is boring and effective. Open your profile as a customer sees it, once a month, and read every field as if you had never seen it. Check the services list line by line and remove what is not true. Check the description. Check the attributes.
This is 10 minutes of work and the highest value AI related task on this list, because it is the only one where you control what the AI produced.
A structured pass over the whole profile is covered in optimizing a Google Business Profile, and the GBP audit tool will flag the fields that drifted.
Why the profile is now doing double duty
The AI features are reading your profile as source material. So every field is now doing two jobs: feeding the ranking system, and feeding the thing that writes your summary.
That raises the value of accuracy specifically, as opposed to completeness for its own sake. A half filled services list used to just be a missed relevance signal. Now it is an invitation for Google to fill the gap itself, with guesses drawn from your reviews and your website.
The primary category still carries more weight than anything else on the profile, and it is the one people most often get wrong. A restaurant that makes its money on pies should be a pizza restaurant, not a generic restaurant, not an Italian restaurant, unless the menu genuinely says otherwise.
If you are unsure what your competitors are set to, the category finder shows the options and what similar businesses use.
One more piece worth knowing: the pages on your own site feed the same picture. A former Googler spelled this out on stage at Whitespark’s Local Search Summit. Google pulls business details off contact and location pages and links them to the profile.
So the address block in your website footer, the hours on your contact page, and the old location page you forgot to delete are all inputs. If a generated description says something wrong about you, the source is often a page you own.
Cleaning up your name, address and phone number across the web is covered in citations and NAP consistency.
How to earn a spot in the listicles AI quotes
If generated answers cite roundups, the practical target is the roundup. There is no trick here, which is why most articles avoid the topic.
- Find what actually ranks. Search the queries a customer would use, in your city, and write down the roundup pages that appear. Best pizza in the city, best deep dish, best late night food, best places for a group. Those pages are your target list.
- Sort them by who accepts submissions. Local publications, neighborhood blogs, food newsletters, tourism boards, and city guides frequently take pitches or updates. National directory listicles usually do not, or they charge.
- Pitch something specific. A pizza restaurant with an actual story, a wood oven, an unusual dough, a family recipe, a late night window, is pitchable. A pizza restaurant with no angle is not, and no amount of outreach fixes that.
- Ask to be corrected, not just added. If a roundup already lists you with the old address or the wrong hours, that is an easy email and a real win, because that stale page may be feeding a generated answer right now.
- Accept the timeline. This is quarters, not weeks, and it is the least predictable work in local SEO. Do not let it displace the profile and review work that pays reliably.
What not to do: buy your way onto a paid listicle and then treat it as a ranking strategy, or spin up your own thin roundup that puts you at number one. The first is expensive with unclear return. The second is the exact pattern that made this problem visible in the first place.
The searches that ignore AI completely
This is the part the AI panic leaves out, and for a restaurant it is most of the business.
A person standing four blocks away, hungry, opening the Maps app and tapping Directions, never sees a generated answer. Neither does the person who taps Call to ask if you can fit a party of eight. Neither does the person filtering for Open Now at 11 at night.
That behavior sits inside Maps, and it is driven by proximity, your category, your hours, your photos, and your reviews.
For a pizza place specifically, calls and direction requests are close to the whole game. Those are the actions that turn into a receipt an hour later. If your reporting tracks generated answer citations but not calls and direction requests, you have built a dashboard that measures the least valuable thing that happened this month.
So the measurement discipline is unchanged. Track pack position across a grid of points around your restaurant, because a single position number hides the fact that you rank first at your own door and eleventh six blocks north.
That is what geo grid rank tracking is for, and the map pack rank checker will show you the spread.
Query by query: where the effort goes
| What the customer types | What the page tends to show | What actually moves it | Worth it for a pizza place? |
|---|---|---|---|
| pizza near me | Pack and Maps dominate, generated answers are rare | Proximity, primary category, reviews, profile completeness | Highest priority. This is the money query |
| best pizza in [city] | Overview more likely, and it tends to quote roundup pages | Being on those roundups, plus pack presence underneath | High, but the work happens on other people’s sites |
| gluten free pizza [neighborhood] | Mixed. Often a pack plus a summary above it | Accurate services and attributes, a matching page on your site | High, and cheap, if you genuinely offer it |
| does [your name] deliver on Sunday | Your panel answers, sometimes with generated text | Hours, attributes, services, and your own contact page | High. 10 minutes of work, direct revenue |
| what is Detroit style pizza | Generated answer usually absorbs it entirely | Good content, if you want brand recall | Low. Do not build a strategy on it |
The pattern in that table is the whole argument. The closer a query sits to buying dinner tonight, the less AI touches it. The further away, the more it does. Prioritize accordingly. For how these elements physically stack up on the page, the local SERP explained maps the layout.
Where AI writing tools help, and where they get you in trouble
Now the other half: AI inside your own workflow. It is genuinely useful for first drafts and genuinely dangerous when it runs unsupervised into a system with content policies.
Review replies
Google now screens every business reply against content policy before it goes live. Most are through in 10 minutes or so, the ceiling is 30 days, and the API marks each reply pending, rejected, or approved.
GMBapi’s review of 12,752 rejections found 92.6 percent attached to five star reviews, with the average rejected reply sent about 50 days behind the review it answered.
Read that second finding again, because it is the practical one. Late replies get rejected. If you batch 100 replies through a generator every other month, you are running exactly the pattern that gets screened out.
Reply within days, keep replies short and specific to what the customer wrote, and never paste a promotional offer into a reply to a five star review.
A review response generator is fine for the first draft of a difficult reply. It is not fine as an unattended queue. More in replying to Google reviews.
Descriptions and services
Writing your own description is worth doing precisely because Google will otherwise generate one. Write it yourself, keep it factual, and mention what you actually sell. A description generator saves you the blank page, but read every sentence before it goes live and cut anything you cannot back up.
Posts
Useful to know before you automate anything. What’s New posts stay on the profile for roughly six months, not the week that older advice keeps repeating. Offer and Event posts run to their end date.
The first 100 characters or so show before the click, and phone numbers in post text are not allowed.
Google shipped native scheduling, multi location publishing, and repeat posts, so you may not need a third party layer at all. If you sell one pizza special a week, one scheduled post a week is the entire program.
What to measure, and what you cannot
Keep this list short and refuse to expand it.
- Pack position across a grid of points, checked on a fixed schedule so the comparison means something.
- Calls and direction requests from the profile, which is the closest thing to revenue that Google hands you free.
- Review count and rating over time, plus how many replies you sent and how fast.
- Whether your target queries show an overview at all, checked by hand once a quarter. That is enough to reprioritize.
What you cannot measure honestly today is your share of generated answers. There is no published methodology and no verifiable count.
An AI visibility checker can tell you whether an assistant currently mentions you for a given prompt, which is a useful spot check. It is a sample, not a metric to set a target against. Be suspicious of any report that puts a confident percentage on it.
A 30 day plan for a pizza restaurant
- Week one, read your own profile as a stranger. Every service, every attribute, the description, the hours, the holiday hours. Delete anything Google added that you do not offer. Fix the primary category if it is vague.
- Week one, fix your own website. Address, phone, and hours on the contact page must match the profile exactly, since Google crawls those pages and associates what it finds with your listing.
- Week two, run a grid check on your three real money terms. Pizza near me, pizza delivery in your neighborhood, and your strongest specialty. Save the baseline before you change anything else.
- Week two, search those same three terms by hand and note which ones return a generated answer and which return a clean pack. That split decides where the next month goes.
- Week three, clear your review reply backlog and get current. Then set a rule: every review answered within a few days, forever. This does more for you than anything AI related on this list.
- Week three, list the roundup pages ranking for best pizza in your city. Pick the three most plausible and send three real emails.
- Week four, one post about something genuinely happening. A special, a new item, a holiday closure.
- Week four, rerun the same grid and compare against your baseline. Then repeat the whole loop next month.
If you want that turned into a repeatable monthly rhythm rather than a one time push, a monthly local SEO plan covers the cadence.
Four claims about AI and local search that are not true
“AI killed local search.” A hungry person opening Maps and tapping Directions has not touched an AI product. For a restaurant, that is the majority of the demand. Something changed, but that is not what changed. Is SEO dead takes the wider version of this argument apart.
“You need an AI optimization service.” There is nothing to buy. The inputs a generated answer reads are your profile, your reviews, your website, and third party pages that mention you. Every one of those was already on the list.
“Add every service the AI suggests.” No. Adding services you do not perform creates policy exposure and produces phone calls you have to decline. The primary category carries more weight than a long tail of padded services, and inaccurate entries make the generated description worse, not better.
“Fill in every service area you can reach.” The leak everyone cites puts service area data on the display and geography side of the ledger, with no ranking role at all. Listing 40 suburbs does not extend your reach, it only makes the listing less accurate about where you work.
FAQ
Does schema markup get me cited in AI answers?
Nobody has published confirmation that it does. Structured data does make your key facts machine readable, which is cheap insurance for a page that lists your hours, address, and menu.
Add it because it is a 10 minute job with no downside, using something like a local business schema generator, not because someone promised you a citation.
Should I block AI crawlers from my website?
For a pizza restaurant, no. Publishers with paywalled archives have a real argument here. You do not. You want machines reading your hours, your menu, and your address, because if they cannot read them from you they will infer them from somewhere older and worse.
Do other assistants use Google’s local data?
At least partly. Microsoft Copilot has been seen putting Google Maps review stars in front of local queries. The practical takeaway is that your Google reviews carry beyond Google, which is another reason the review program is the most useful AI work available to you, even though nothing about it feels like AI work.
A competitor’s panel lists services they do not offer. Should I report it?
Probably it was generated rather than entered, so it is not evidence of anything sneaky. You can suggest an edit, but it rarely changes a ranking and it costs you attention. Spend the same 20 minutes auditing your own profile, where the same thing is likely happening to you.
Do AI written blog posts hurt my local rankings?
Not directly, but they rarely help either. On page work is roughly 15 percent of what decides the pack and roughly 33 percent of local organic. Generic generated posts do not move either number, because they say nothing a local reader needs.
A real page about your delivery zone, written by someone who knows the neighborhoods, outperforms 30 generated posts. That is the argument in local SEO content strategy.
How often should I check the AI generated parts of my profile?
Monthly is enough for most restaurants, plus immediately after any menu change, hours change, or burst of reviews. Those are the moments when Google has new material to summarize, so those are the moments new details tend to appear.
Where ReviewResponseTool fits
ReviewResponseTool handles the measurement side of all this. It tracks your pack position across a grid of points around the restaurant, so you can see where you actually appear rather than a single flattering number.
It also audits the profile fields that AI features read from, and monitors reviews so replies go out while they still get approved.
Pricing runs $16 a month for a single location, a flat $37 once you are running two to seven, and $5.33 per location per month from eight locations up. Pay yearly and 10 percent comes off. The trial lasts three days and skips the card entirely.
You can start the trial and have a baseline grid inside an hour.
The thing it cannot deliver is a citation inside an AI Overview, and neither can anything else on sale today, because no published mechanism exists to buy or verify placement in a generated answer. Anyone claiming otherwise is guessing with a confident voice.
The parts you can control are the profile, the reviews, the site, and the roundups that mention you. Those are the parts worth paying for.