Home-Backup Storage: From Brand Interest to AI Alternative Recommendations

Gravity industry case 01 · Clean energy and home storage, North America | 65 buying-task questions · T0: 138 answers on the original 46 questions · T1: 585 answers on all 65

1. The Challenge: No Standard Engineering Example When Named, and Difficulty Entering Unbranded Shortlists

In North American home-backup and whole-home storage, where purchases commonly cost several thousand to tens of thousands of dollars, consumer questions to ChatGPT, Perplexity, and Gemini fall into two distinct groups:

1. Unbranded category-shopping and competitor-alternative questions: “Which alternatives to Tesla Powerwall should an American household evaluate to maximize rooftop-solar self-consumption?” “Which modular systems let me start at 5 kWh and expand later?” “How should a hurricane-prone household back up its refrigerator, internet, and central air conditioning?” At T0, the brand had 0 mentions, 0 recommendations, and 0 official-site citations on commercially valuable questions such as Powerwall alternatives. Gemini shortlisted it on 0 of all 22 natural-recommendation questions.
2. Branded deep specification and engineering-boundary checks: buyers who had shortlisted the flagship modular-backup system or the new 5–30 kWh stepped-storage system asked how to calculate real runtime after temperature, conversion losses, and the SOC window; how it couples with existing rooftop solar; and how to distinguish indoor/outdoor installation ratings and tax-incentive deadlines. Without structured calculations and engineering FAQs, AI turned to third-party forums and could confuse submodel power figures such as 4 kW and 7.2 kW.

2. Gravity's Work: Deriving Content and Fact Alignment from Real Buying Tasks

Following the principle of starting with real buying questions and connecting product facts, official content, external evidence, and AI answers to find recommendation gaps, Gravity worked with the client team on three steps:

1. Built 22 buyer Demand Cells and a two-module evaluation baseline. The North American home-storage decision journey was mapped into 22 demand units, covering modular expansion, Powerwall alternatives, EV charging during outages, multi-day hurricane backup, AC/DC coupling retrofits, millisecond transfer, UL 9540 safety compliance, and more. The assessment keeps the original 46-question module—35 unbranded natural questions plus 11 branded diagnostic questions—separate from 19 client-added new-product pain-point questions.
2. Published highly structured calculation and comparison assets.

• Home-Backup Runtime Estimation Guide: instead of vague “up to several days” language, the official site published the full calculation chain: 12.288 kWh nominal capacity → 100% to 10% SOC window, 0.90 → 95% remaining capacity → 0.90 temperature factor and 90% AC conversion efficiency → 8.51 kWh usable AC energy → approximately 20.8 hours with a 9.84 kWh/day, roughly 410 W, load and no recharge. The guide clearly states that these are planning estimates, not guarantees.
• Modular-start and competitor-alternative comparisons: dedicated content addresses starting at 5 kWh and expanding up to 30 kWh, comparing Powerwall alternatives, and adding storage to existing rooftop solar. Power, expansion limits, and suitable household profiles are organized for straightforward AI extraction.

3. Completed 585 retest answers and 93 scoped fact checks. After launch, T1 covered 65 questions × 3 platforms × 3 independent same-day rounds = 585 answers, 580 with verifiable source links. Ninety-three claims involving power, price, compatibility, installation, and tax incentives were checked individually against official first-party evidence.

3. Observed Results: A Breakthrough in Natural Shopping and Stronger Official-Source Authority

3.1 Entering a Leading Powerwall-Alternative List; Gemini's Natural-Question Citation Rate More Than Doubled to First Place

3.2 Branded and Model Checks: Citation Share Gained More Than 20 Points and the Engineering Example Was Reproduced Verbatim

• The original 11 branded diagnostic questions: direct official-site citation rate rose from 81.82%, 27/33, to 90.91%, 30/33, in the complete sample. Excluding one generic regulatory question that did not contain a brand name, citation reached 100%. Official-site citation share rose +20.44 percentage points, from 43.12%, 116/269, to 63.56%, 150/236. Perplexity's official-link share increased from 37.56% to 76.36%.
• The published runtime guide completed a citation-and-adoption loop: it received 4 direct links on core runtime diagnostic questions. In all 3 rounds, Perplexity reproduced the delivered calculation chain: > “12.288 kWh, 100% to 10% SOC, 95% remaining capacity, a 0.90 temperature factor, and 90% discharge-to-AC efficiency—the available energy is 8.51 kWh; with a 9.84 kWh/day load and no recharge, estimated runtime is about 20.8 hours. These are planning estimates, not guarantees...”
• The 19 client-added new-product questions: all 171 retest answers carried verifiable sources. Average brand mention rate reached 97.66%, including 100% on Gemini. Average official-site citation rate was 87.72%. Perplexity cited the official site on 19/19 questions in every round, and ChatGPT also reached 19/19 in round one. More than 80.51% of all reference links came directly from official brand domains. This added module is reported separately from the original baseline group.

3.3 Retesting Guides the Next Iteration: 93 Fact Checks Identify Five Evidence Tasks

The 93 scoped checks covered 88 specific claims plus 5 entries making no specific claim. They confirmed 48.86%, 43/88, supported accurately by official first-party evidence. Only 3.41%, 3/88, involved model-power confusion or outdated tax-incentive dates. The checks also identified two concrete growth levers:

1. Bridge the gap between “cited and listed” and an explicit recommendation. Each retest round already contained 23–28 official-site citations on natural questions. About 30% included the brand in a shortlist with neutral wording such as “Here are some examples.” A Best For / Decision Matrix table with clear suitable-audience qualifiers on already well-cited blogs offers a concrete way to strengthen AI's basis for explicit recommendations; its effect remains a target for the next retest, not a guaranteed outcome.
2. Fill five precise public-evidence gaps. For 42 claims awaiting additional public evidence, the next tasks are: Installation Clearances and Outdoor-Panel Wall-Penetration FAQ; Cross-Generation Accessories and AFCI Breaker Compatibility White Paper; Standard Package Configurations and Retail Price List; Individual Battery-Module Replacement Policy; and Offline Local Operation and Privacy Boundaries. Together they form a “find the gap → add evidence → retest” growth cycle.

4. Further Reading

• Industry overview: AI Is Starting to Make Buying Decisions: GEO Fieldwork Across 10 Brands →
• Next case: Smart ergonomic office furniture: turning model differences into AI-usable buying criteria →

Discuss your industry's GEO opportunity with Gravity →