We used open records to look up over 1,000 complaints about Georgia Power to the Georgia Attorney General's Consumer Protection Division. Here's what we found:

You can access all the files using this link.

Referred and Closed

What 1,082 Georgians told the state about their utility service — and what the state did about it.
Source Ga. Dept. of Law, Consumer Protection DivisionPeriod Jan 2020 – Dec 2025Records 1,082Pages 3,904

These are not PSC complaints

The documents are intake records from the Georgia Department of Law's Consumer Protection Division — the Attorney General's consumer arm. Each one is a "workflow": a person contacted the state about their utility, a counselor wrote up what they said, and the file was closed with a disposition code.

That distinction matters for how you can use this. Nobody in this dataset necessarily got a hearing, a docket number, or a regulator's attention. What they got, overwhelmingly, was a phone number.

Georgia Power is the subject of roughly 95% of the records. The remainder name gas marketers, EMCs, municipal utilities, or a company impersonating Georgia Power. This is, in practice, a file on one company.

The single most common event in this dataset is a Georgian describing a problem to the state and being told to call someone else.

Two

Out of 1,082 records spanning six years, the number in which the file shows the consumer's problem actually being fixed is two.

1,082 records, Jan 2020 – Dec 2025

855 closed by referral to the PSC (79%)

511 closed the same day they opened (47%)

2 show a resolution for the consumer

Figure 1

What happened to 1,082 complaints

Disposition codes come from the source records; outcome was coded by reading each file.

1,082 Contacts the state took in

948 Referred somewhere else and closed 87.6%

98 Given information, no action 9.1%

25 Dead end — consumer unreachable or file insufficient 2.3%

9 File shows the company engaged 0.8%

2 File shows the problem resolved 0.2%

"Referred elsewhere" combines referrals to the PSC, to the company's corporate office, and to other agencies. "Company contacted" means the file shows the business engaging with the complaint; it does not mean the consumer got what they asked for.

Volume quadrupled, and it is not evenly spread

Complaints rose from 70 in 2020 to 361 in 2025 — with the sharpest jump between 2023 and 2024. Two spikes stand out: July–September 2024 (138 complaints in three months, the heaviest stretch in the file) and January–March 2025 (152).

The seasonality is consistent and unsurprising once you see it: August is the single biggest month across all six years, January and February close behind. Summer cooling bills and winter heating bills are what drive people to call the state.

What people were actually calling about

Money dominates. High bills, billing errors, and rate-plan problems together account for 32% of the file. Add payment processing and disconnection and it is more than half.

The mix also shifted. Comparing 2020–2022 with 2024–2025, "high bills / bill shock" went from 4.7% of complaints to 15.6% — more than tripling as a share of the file, on top of a fourfold increase in total volume. Complaints objecting to rates as set, including fuel cost recovery and nuclear construction cost recovery, more than doubled as a share over the same span.

Figure 2

Complaints by primary category. Each record was assigned one primary category.

High bills / bill shock    140

Billing errors & disputed charges   120

Rate plan problems    92

Other     88

Outages & reliability      81

Payment processing & arrangements     75

Disconnection & reconnection     70

Property damage & safety     60

Metering      56

Customer service & account access      53

Account transfer, landlord-tenant & deceased      46

Rate case, fuel costs & policy objections      43

Fraud, scams & identity theft      35

Vegetation management & easements      34

Service establishment & construction     32

Deposits & credit requirements       21

Unclear / illegible      12

Solar, net metering & distributed energy     9

Affordability & hardship      8

Energy efficiency & assistance programs      7

The stories that hold up

153 records carry a vivid, specific, first-person account with a clear chain of events. Of those, 94 have a quote verified character-for-character against the source document and no detected personal information — those are the safest to work with, and they are broken out on their own tab in the database.

The strongest cluster is disconnection during weather extremes, often with a medical or child-safety dimension the company had been told about.

Disconnection Batch 4, p.361–364 Aug 7, 2024 Severity 5

"I don't have no money until I get paid and now I'm stuck in the heat. With my child."

Power cut in 95-degree heat one day after her payment-arrangement date, over a balance of roughly $100.

Disconnection Batch 3, p.719–721 Jan 26, 2024 Severity 5

"I felt like I was dying and all over again I experienced this through the pandemic already and I know that as a human being I should have some rights"

Consumer on a 24-hour breathing machine says service was cut over a past-due balance despite a submitted doctor's statement.

Rate plan Batch 5, p.277–280 Jan 30, 2025 Severity 5

"Electricity is not a luxury-it is a necessity."

A senior on a fixed-rate plan she had held for 19 years was removed from it over late payments tied to her Social Security deposit date, and now faces $1,737.96 and disconnection.

Service establishment Batch 3, p.119–124 Mar 14, 2023 Severity 5

"me and my babies were in the cold no hot water no food no heat all because of the employees Kelvin refuse to give me service intentionally to give me a hard time"

Refused service, pushed onto prepaid despite stated policy, and charged about $773 of unverified old debt.

Quotes are reproduced exactly as they appear in the record, including original spelling. Each is traceable to a batch and page number in the database.

Patterns, not just bad luck

389 records show something a regulator or intervenor could act on as a practice rather than an individual dispute. Five clusters recur often enough to be worth naming

Another person's debt as a condition of service

Consumers repeatedly describe being required to pay a prior tenant's, a landlord's, or a deceased relative's balance before service would be connected in their own name. One record involves an 80-year-old landlord disconnected and told to clear her absconded tenant's $3,000 balance before anyone could have service at the address.

Payment arrangements refused, then disconnection

A recurring sequence: consumer requests a due-date change or installment plan, is refused and told the full balance is due, is referred to a charity, and is disconnected. In several files the disconnection follows within a day of a payment-arrangement date the consumer believed they had met.

Disconnection where the vulnerability was known

162 records name a specific vulnerability — elderly (64), disabled (45), children in the home (42), fixed income (23), medical equipment (15). Thirty of those are disconnection complaints, ten involving plug-in medical equipment the company had reportedly been told about.

Rate plans that cost more than the default

Consumers describe being enrolled in or steered toward Smart Usage, flat-rate, or prepaid plans and later discovering the standard residential rate would have been cheaper — sometimes learning it from the company's own representative. 92 records fall in this category.

Large back-bills after months of paying as billed

Files describe year-end or multi-month catch-up bills issued to consumers who had paid every statement as it arrived. The largest single back-bill in the file is $35,000; a meter-tampering allegation carries a $65,000 demand.

On the dollar figures

428 records state a dollar amount. The median is $421 and 80 records claim $1,000 or more. One record — a multi-party claim alleging disconnection of disabled customers during freezing weather — states $800,000,000; it will wreck any average, so use the median.

A small but specific file

46 records touch clean energy, efficiency, or how generation choices land on bills. It is one of the smallest of the lenses, and the honest read is that this dataset is not primarily an energy-policy file — it is a billing-and-shutoff file. But two threads are worth pulling.

Solar interconnection delay, 2021–2022

A cluster of records — from homeowners and from installers — describes rooftop solar sitting idle for months awaiting interconnection approval or a meter change. One installer reports reviews running months past a stated 15-day estimate; another alleges the company withholds customer usage data from solar companies. Individually these are service complaints; together they are an interconnection-queue story with dates.

Riders and cost recovery, itemized by the ratepayer

Consumers object, by name and by dollar figure, to environmental compliance cost recovery and nuclear construction cost recovery appearing on their bills — one itemizes $44.74 in such charges on a $181 bill and asks why. Complaints in this category more than doubled as a share of the file between 2020–2022 and 2024–2025.

Also present: EV rate metering disputes, an efficiency-program rebate never paid, and an LED retrofit sold door-to-door that raised the tenant's bill. Explicit data-center complaints are essentially absent from this window.

Three things about the source material:

The production contains a duplicate batch

"Batch 7 2.pdf" is a duplicate export of "Batch 6.pdf" — identical page count, and all 146 of its intake IDs already appear in Batch 6.

Half the corpus came through OCR

1,865 of 3,904 pages had no text layer and were optical-character-recognized. They are typed documents, not handwriting, so accuracy is high — but it is not perfect. Every quote used in this memo was matched character-for-character against the extracted text, and the database marks which quotes passed that check. Two of 571 offered quotes failed and are marked accordingly.

Method

Seven PDF batches (3,904 pages) were processed in the numerical order of their filenames, which preserves the production sequence. Pages carrying a native text layer were extracted directly; the 1,865 scanned pages were rendered at 200 dpi and OCR'd. Records were split on the intake-workflow header rather than the "*Intake" title line, because OCR loses that title on scanned pages — using it alone would have found 10 records in Batch 4 instead of the 229 actually there.

Dates, disposition codes, intake IDs, and page references come directly from the source documents and were never model-generated. Categories, severity, vulnerability, resolution state, and the three flags were assigned by reading each record against a written coding guide; they are judgments and should be spot-checked before load-bearing use. Quotes were verified programmatically against the extracted text. PII detection is deterministic pattern-matching filtered by corpus frequency, so that letterhead boilerplate — Georgia Power's own address, the PSC's phone number — is not mistaken for a consumer's details; it is a screening tool for review, not a legal determination.

Companion file: Georgia_Utility_Complaints_Database.xlsx — all 1,082 records with every coded field, plus "Ready to Quote" and "PII Review" tabs.