Review Management at Scale: How to Run 1,200 Locations Across 200 Businesses

Two hundred client businesses averaging six locations each is twelve hundred locations. At a median velocity of roughly three new Google reviews per location per month, that portfolio takes in about 3,600 reviews a month and 43,200 a year, sitting on top of a historical backlog that is comfortably into six figures. Nobody reads that. The instinct is to hire, or to automate every reply and hope nobody notices. Both are the wrong move, because past a few dozen locations review management stops being a writing task and becomes a routing task.

What follows is how that routing works in ReviewMankey, the AI reputation platform we build at Dude Lemon, using the real product screens rather than mockups, and the 2026 research that says why it now matters more than it did last year.

What “at scale” actually means, in numbers

Review management gets discussed in adjectives. It is worth doing in arithmetic, because the arithmetic is what decides whether a workflow survives the fiftieth client. Here is the portfolio in this article, with every derived figure marked as derived.

Now price the labour. Assume three minutes to read a review, decide the response, write it in the client’s voice and publish it. That is generous for a five-star review and optimistic for an angry one. At 3,600 reviews a month it is 10,800 minutes, or 180 hours, or a little over one full-time person doing nothing else. At restaurant velocity it is 373 hours, more than two people. That is before escalations, before client reporting, and before anyone logs into two hundred separate platform accounts.

Why the stakes went up in 2026

Two independent 2026 datasets and one practitioner survey moved the goalposts in the same direction. Consumers now read reviews more often, demand fresher ones, expect a reply within a day, and increasingly ask an AI assistant rather than a search engine. Meanwhile the average business still leaves more than half of its Google reviews unanswered.

Forty-two percent of consumers say they are unlikely to use a business that never replies. The average business does not reply to 53% of its Google reviews. Nineteen percent of consumers now expect a reply on the day they post, up from six percent a year earlier, while one vendor benchmark built from platform data between January 2025 and February 2026 puts the average time to respond to a Google review at 2.7 days. Every one of those gaps is an operations problem, not a copywriting problem.

What a rating point is actually worth

The academic literature on this is older than the software, and it is unusually clean. Three findings are worth knowing before you argue for a budget.

One inbox, and one number per business

The screen above is a single account: 1,247 reviews under management, 4.3 stars lifetime, 4.1 over the last thirty days, an 87% response rate, five open actions, and a reputation score of 72 that is up three points on the previous period. The period selector runs 7 days, 30 days, 90 days and 12 months, and the sync indicator is live rather than a nightly batch.

The pair worth staring at is 4.3 lifetime against 4.1 for the last thirty days. A lifetime average is a slow-moving number that hides a live problem for months, because a business with a thousand historical reviews needs an enormous run of bad ones to shift it. The thirty-day figure moves immediately. Whitespark’s 2026 report puts review recency among the highest-weighted individual review factors, which is exactly why a lifetime average is the wrong number to steer by and the right number to report.

At portfolio scale the reputation score does the triage on the dashboards themselves. You cannot read two hundred dashboards a week. You can read two hundred scores, sorted by change, and open the six that moved.

Every source, normalized into one sentiment split

Google Business Profile at 4.3 across 612 reviews, the Play Store 4.1 across 142, the App Store 4.5 across 121, and first-party website reviews 4.6 across 78. Normalizing the payload once at ingest, so that a rating, an author, a timestamp, a platform and a location mean the same thing regardless of where the review came from, is what allows a single queue, a single rule engine and a single metric set to work across sources that otherwise share nothing.

It also surfaces spreads that a blended average erases. There is half a star between the Play Store and the website on this account. An app-store rating problem and a storefront rating problem have different owners and different fixes, and a single blended average tells you about neither. Most platforms in this category cover storefronts or app stores, not both; ReviewMankey’s write-up of managing locations and apps in one place explains why that split exists.

Sentiment sits on its own axis, not derived from the star rating: 57.7% positive, 24.9% neutral, 11.6% negative and 5.8% very negative. That last pair is 217 reviews, and it is the only part of the 1,247 that genuinely needs a person. Separating sentiment from rating matters because a three-star review with a specific, fixable complaint is worth more attention than a one-star review with no text, and the star count cannot tell them apart.

A review is a unit of work, and it has a state

Reviews move across a board: Pending, In Progress, Replied. Cards carry the sentiment the model assigned, the workflow tags a human added, the location and the date, and they filter by platform and by star rating. The drawer on the right opens the whole record: column, sentiment, response state, rating, platform, received timestamp, tags such as VIP and Positive PR, the reviewer, and the full response history.

Making state explicit is what turns response coverage from an estimate into a measurement. The 87% response rate on the dashboard is a count of records in a state, not a guess, which is the only reason it is safe to put in front of a client next to the industry average of 46.9%.

The mechanic that makes hundreds of businesses possible: policy per location

Platforms connect at the account level: one Google account here holds eight locations, another holds four, and the Play Store and App Store connections sit beside them with their own last-sync timestamps. Open the Google drawer and you get accounts, then a searchable and paginated list of all twelve locations.

The badges on each location are the whole point. Reply policy is a property of the location, not the account, not the client, and not the workspace.

This is the difference between software that manages a business and software that manages a portfolio. If reply policy were an account-level setting, onboarding one nervous client would mean either dropping the whole portfolio to drafts-only or asking that client to trust automation on day one. Because policy lives on the location, the risk appetite of one client never constrains another, and a single underperforming branch can be pulled back to drafts-only without touching its eleven siblings.

The onboarding arithmetic follows from the same design. A Google account holding eight locations connects once and brings eight locations with it. Across a portfolio averaging six locations per client, the unit of setup work is the account rather than the location, which is roughly a sixfold reduction in connection steps. What remains is a policy decision per location, which is a judgement call measured in seconds, not an integration measured in hours. For the single-client version of this, ReviewMankey’s own guide to multi-location Google review management covers the same ground at one brand rather than two hundred.

Rules that triage before a human looks

Rules are priority-ordered from P1 down, each individually enabled or paused, each with conditions on the left and consequences on the right. P1, urgent safety concern, matches any platform and any location at one to two stars where sentiment reads urgent or angry and the text contains unsafe, injury or hazard, and assigns critical severity. P2, refund request signal, narrows to Google Business and two specific locations at one to three stars with frustrated or disappointed sentiment and the keywords refund, charged or money back. P3 catches one-star angry reviews. P4 is the low-rating catch-all. P8, potential spam or fake review, sits paused while its thresholds are recalibrated, which is a state a rule engine needs to have.

1P1  Urgent safety concern                     enabled
2    when
3      platform    any
4      location    any
5      rating      1..2
6      sentiment   urgent | angry
7      keywords    unsafe, injury, hazard
8    then
9      severity    critical
10    observed
11      3 matches in the last 7 days
12 
13P4  Low-rating monitoring                     enabled
14    when
15      platform    any
16      location    any
17      rating      1..2
18    then
19      severity    medium
20    observed
21      27 matches in the last 7 days

The match counts are the interesting column. Four enabled rules produced 49 matches in seven days on this account: 27 in the medium catch-all, 11 on refunds, 8 on one-star anger, and 3 critical. That is the real size of the human queue, and it is a list a person can work in an afternoon. The 1,247 reviews never had to be read; the rules read them.

Ordering matters because the rules overlap deliberately. A one-star review mentioning an injury matches both P1 and P4, and priority decides that it inherits critical rather than medium. This is why the interface is an ordered list rather than a flat set of switches, and why the specific rule has to sit above the catch-all.

Test Rules and View Log are the other half. A rule engine you cannot dry-run against history is a rule engine nobody will touch, which means thresholds calcify at whatever they were on day one. Being able to test a proposed rule against back history before enabling it, and then read a log of what actually fired, is what lets one operator tune escalation for two hundred different businesses without breaking any of them.

Writing the rules themselves is a separate exercise, and the severity tiers, owners and SLAs behind them are worth getting right before you enable anything. The review escalation matrix playbook on the ReviewMankey blog is the long version.

A 4.5 means nothing without the market around it

This location holds 4.5 stars, which is 0.2 above the average competitor and trending up 3%, on 847 reviews trending up 8%. It still ranks third of six, and the review-share donut explains why: the leader holds 35% of review volume in the market, this location holds 14%. The comparison refreshes every six hours.

Rating and volume are different problems with different interventions. Rating is a service-and-response problem measured in quarters. Volume is a review-request problem measured in weeks. A per-location market rank across 1,200 locations tells you which of them is fighting for stars and which is fighting for share, and stops you prescribing the same campaign to both.

Reviews are now the input to AI recommendations, not just search

The most consequential 2026 dataset for anyone running a review programme is SOCi’s Local Visibility Index, which covered 2,751 brands and roughly 350,000 locations. Brands appear in Google’s local three-pack 35.9% of the time on average. Only 1.2% of locations are recommended by ChatGPT, and 11.0% by Gemini. Recommended locations average 4.3 stars on ChatGPT, 4.2 on Perplexity and 3.9 on Gemini.

Those averages describe a filter rather than a ranking factor. Fall below roughly four stars and many AI systems stop considering a location at all, no matter what else is true about it. Above the threshold, volume and recency start deciding. And the audience is no longer marginal: 45% of US consumers told BrightLocal they had used a generative AI tool for a local business recommendation, up from 6% the year before.

There is a second-order effect worth naming. When an assistant answers a question about a local business, it reads what is publicly written, and that includes your replies. Reply text stops being customer service and becomes indexable copy about how the business handles problems. A generic template repeated four hundred times reads exactly as what it is. If you want the reviews you earn to also be visible to search engines and AI assistants, that is the job SearchLift does, and our guide to generative engine optimization covers the mechanics.

The operating model, week one to steady state

• Week one, connect at the account level. One platform account brings every location it owns. Do the same for the app stores and the review sites the client actually cares about, then let history backfill before you change anything. You are buying a baseline, not a starting point.
• Week one, set every location to drafts only. Nothing publishes automatically on a portfolio nobody has read yet. This is not caution theatre. It is how you learn a client’s voice from their real reviews rather than from a brand guideline document.
• Week two, write the rules once and reuse them. A safety rule, a refund rule, a one-star rule and a low-rating catch-all cover most of what matters. Dry-run each against back history before enabling it, and read the match counts honestly: a rule that fires two hundred times a week is not an escalation rule, it is a report.
• Week three, promote the easy locations to auto-reply at five stars. Five-star replies are high volume and near-zero risk, and they are where response coverage moves fastest. This single step usually closes most of the gap to a defensible response rate.
• Month two, promote settled locations to auto-reply at four stars and above. Leave three stars and below with a person permanently. The three-star review is where the actual information lives, and it is the one a template will always get wrong.
• Steady state, work the exception queue and the outliers. One score per business, sorted by change. Open the ones that moved, read the critical escalations, and leave the rest running. If your week is spent inside individual reviews at this stage, the rules are mistuned.

What to measure, and what each number is for

Where configuration ends and engineering starts

Most of what is above is configuration, and that is the point: a portfolio should not need a build. Four things reliably do. Routing critical escalations into a ticketing or on-call system a client already runs, rather than into email. Pushing review sentiment into a warehouse so it can sit beside sales and staffing data, which is where the causal questions actually get answered. Per-brand voice in a group that owns several distinct brands. And reconciling a client’s own location hierarchy with the messier one the platforms hold, which is the single most common reason a location count does not tie out.

That work is what we build, on the same platform rather than beside it.

Frequently asked questions

The short version

None of this makes review management easy at scale. It makes it bounded. Once a review is a unit of work with an explicit state, once reply policy lives on the location rather than the account, and once ordered rules do the triage, adding the two hundred and first business changes the numbers on a dashboard and nothing else about the job. That is the whole design brief.

ReviewMankey is the platform. If you want the reviews you earn to be visible to search engines and AI assistants as well as to customers, SearchLift is the other half of the problem. And if your portfolio needs review data routed into systems you already run, talk to our team.

Sources

• BrightLocal, Local Consumer Review Survey 2026. 1,002 US adults surveyed in February 2026.
• SOCi, Local Visibility Benchmarks for Multi-Location Brands, 2026 Local Visibility Index. 2,751 brands and approximately 350,000 locations.
• Whitespark, 2026 Local Search Ranking Factors. Survey of 47 local SEO practitioners.
• Davide Proserpio and Georgios Zervas, Online Reputation Management: Estimating the Impact of Management Responses on Consumer Reviews, Marketing Science 36(5), 2017. Summarised in Harvard Business Review.
• Michael Luca, Reviews, Reputation, and Revenue: The Case of Yelp.com, Harvard Business School NOM Unit Working Paper 12-016.
• Chris Anderson, The Impact of Social Media on Lodging Performance, Cornell Hospitality Report.
• WebFX, 2026 Google Business Profile Benchmarks, for review velocity and top-three review count reference points.