Ameresco’s AI search problem was one of memory. Chatbots knew the company well enough; they described the business it used to run.

When Leila Dillon, Ameresco’s senior vice-president and chief marketing officer, put 50 questions that prospective customers and investors might ask to ChatGPT and Claude in January 2025, the answers mostly described a cleantech and renewables firm, and sometimes surfaced small, dated projects. She told Business Insider the results were “horrifying”. The audit, run with the PR agency Greenough, produced 47 areas to fix. Ameresco now says it appears in 61% of the answers to those questions in 2026, against 43% in 2025, and that organic traffic to its website rose 50% in the 12 months to June 2026, according to a CMO Insider report Business Insider published on 21 October.

The case stands out because most of what Ameresco changed was not technical. The work sat in positioning, press relations, award entries and content for specific buyers, the off-site and earned signals that independent research increasingly links to AI citation. It is also one of the few company-level generative engine optimisation (GEO) programmes described in public with before-and-after figures, which makes it both a usable template and a fair test of how much such figures can prove.

What did Ameresco find when it audited chatbot answers?

Ameresco found its chatbot profile lagging well behind the strategy it now presents to customers and investors. The Massachusetts-based group, listed on the New York Stock Exchange as AMRC, has been recasting itself around power infrastructure and building upgrades for customers such as data centres, utilities, government bodies and schools. The chatbots mostly called it a cleantech or renewable energy business.

Part of that picture came from Ameresco’s own copy. Its press-release boilerplate through 2024 opened by calling it “a leading cleantech integrator”, and that text ran on its releases distributed through Business Wire. A model asked “What are the top energy infrastructure providers in the US?”, one of the 50 prompts, had little reason to put the company on its list.

Ameresco is not unusual in this. A Semrush survey of US marketing professionals conducted in April found 30% said AI search tools had described their brands inaccurately. An Interactive Advertising Bureau survey in January found 73% of US brand and agency buyers planned to spend more time or resources in 2026 optimising content for AI-generated answers. Business Insider reported both figures without sample sizes.

What did Ameresco change?

Ameresco changed the description first. The boilerplate now calls it “a leading energy infrastructure company”, the same phrase that opens its corporate website, which lists two businesses: Power Infrastructure, and Buildings and Public Infrastructure. A new corporate identity built on those two pillars followed in July 2026.

The rest of the programme ran through channels marketing and communications teams already control. Ameresco redirected award submissions towards power infrastructure projects and announced wins consistently through wire-distributed press releases and industry trade titles. It added question-and-answer content intended to be easier for chatbots to interpret, built material for specific buyers such as data-centre operators and school facilities managers rather than leaning on general corporate copy, and started a series on customer outcomes. It now re-runs the 50 prompts every quarter, compares its showing against competitors, trains its marketing team and shares findings internally each week.

The published account describes no schema markup, llms.txt files or other technical work. That is consistent with Google’s guidance of 15 May, which told site owners that optimising for AI Overviews and AI Mode is still SEO and that most specialist hacks are unnecessary. It is also consistent with a Walker Sands benchmark of 828 enterprise B2B companies, which found the median brand cited in 3% of relevant AI Overviews, alongside Ahrefs data linking AI visibility more closely to brand mentions than to backlinks. Dillon was clear that the job does not finish: she said chatbot answers shift from one week to the next and warned against updating a website and assuming the work is done.

How much do the 61% and 50% figures prove?

The figures show which way things moved, but not why. Both come from Ameresco, and neither appears in a press release, investor presentation or filing. The 61% is the share of answers to Ameresco’s own 50 questions in which the company appeared. Ameresco has not published the questions, the number of runs per question, the model versions tested each year or how an appearance was scored. Chatbot answers vary between runs and across model updates, so a self-selected prompt set checked once a quarter is a tracking tool rather than a controlled experiment.

That is also why the 61% cannot be set against the Walker Sands 3%. Walker Sands measured how often a company’s domain was cited across millions of relevant Google searches. Ameresco measured whether its name appeared in answers to questions it wrote about its own market.

The traffic gain is looser still. Organic visits rose 50% over a period that included a corporate repositioning, a year of award and press activity and a brand launch, and Dillon told Business Insider it was too early to say how the gains translate into business results. A critical survey of 45 GEO studies published in July found no technique that reliably lifts AI visibility across engines, and showed how growth in AI platforms themselves can be mistaken for an optimisation gain. Nothing Ameresco has disclosed rules that out. Its budget, headcount and agency fees have not been disclosed either.

For a B2B software company the transferable part is the process: list the questions buyers and investors actually ask, test them in more than one assistant, record what is wrong or missing, and correct the description everywhere it is syndicated before rewriting individual pages.

Ameresco reports third-quarter results on 9 November. The 50 questions behind its 61% remain unpublished.