It is the Tuesday before the public meeting on a new compressor station. A woman who lives three hours away in the city, whose mother owns the pasture next to the proposed site, opens ChatGPT on her phone and types your company’s name and the question she has been chewing on for a week: “Are they a safe operator? Have they had leaks?”
She will not read the docket. She will never open the environmental reports your consultants spent a year on. She reads the answer on her screen, takes a screenshot, and texts it to her mother and two of the neighbors. By Thursday that screenshot may be the most widely read thing anyone on that road has seen about your project.
The obvious response is that none of this matters, because the permit gets decided on the record. That is correct. I’m not going to argue with it. Regulators decide on filings, not on what a chatbot told somebody’s daughter. But the record is not the only thing a project has to get through. Turnout at the meeting, the tone of the comments that get filed afterward, whether the local paper calls you before it runs the story, whether the landowner signs the easement or calls a lawyer: those are all shaped before anyone opens a docket, and more of that shaping now starts with a question typed into a chat window.
The person asking is rarely the landowner
Here is a fact that cuts against my own argument. If the landowner is in their seventies, they are among the people least likely to ask. In Pew Research Center’s February 2026 survey, 19% of Americans 65 and older said they use ChatGPT. Among adults under 30 it was 61%, and among those 30 to 49 it was 55%. Across all U.S. adults it was 44%, up from 18% in 2023.
So the person asking is the son who handles his parents’ paperwork from another city. It is the reporter at the regional paper who has forty minutes before deadline and has never heard of your company. It is the aide to a county commissioner who got asked about the project after church. And ChatGPT is only one door: Pew also found that about half of U.S. adults now use AI chatbots such as ChatGPT, Gemini or Copilot. None of these people are your opponents. Most of them have simply never had a reason to look you up before.
What the answer is made of
The researchers who coined the term generative engine optimization described these systems in one line: they answer a question by “synthesizing information from multiple sources and summarizing them,” and the people who publish those sources have “little to no control over when and how their content is displayed.” The second half is the part to sit with. Your company does not choose which pages the answer leans on.
Think about what is easy to find on a project like this. A news story about a release at a different company’s facility in the same county. An opposition group’s FAQ, written in exactly the plain questions a worried neighbor types. A two-year-old article about a lawsuit that was later dismissed, with no follow-up story about the dismissal. An old enforcement notice filed under a company name close enough to yours to get blended in. Your own account of the project exists too, but it lives in filings written for a regulator, organized by resource category and answering the questions an agency asks rather than the ones a neighbor asks.
I can’t tell you how often a chatbot gets a particular project wrong, and I’d be wary of anyone who quotes you a number for it. What I can tell you is that when nobody has published a clear answer to the neighbor’s question, the machine answers anyway. It builds the answer from whatever is lying around.
The people typing those questions are not one audience, and they are not deciding the same thing:
| Who is asking | What they are deciding | What the answer tends to be built from | What you can put in front of it |
|---|---|---|---|
| A landowner’s adult child | Whether to tell a parent to sign, hold out, or hire a lawyer | Old incident coverage, opposition FAQs, companies with similar names | A plain-language project page that answers safety, noise, truck traffic and who-to-call questions in the words people actually type |
| A local reporter on deadline | How to describe your company in the first paragraph | Whatever ranks first for your name plus “leak,” “fine” or “lawsuit” | A dated record of past incidents and how each one was resolved, on a page you control |
| A county official or their staff | Whether to support, stay quiet, or ask for conditions | Neighbors’ comments, news coverage, general statements about the industry | The docket number, sourced local facts about the project, and a named person to call |
| Someone from town applying for a job on the project | Whether working for you is a good bet | Review sites and any layoff or safety coverage | Accurate pages about the local work: the roles, the training, and who runs the site |
Can you change the answer before the meeting?
Partly. You cannot edit what a model says, and anyone who promises you control over it is selling something. You can change what it has to work with, and on a single project the useful moves are smaller than most people expect.
Ask the questions yourself first. A week or two before the meeting, have someone outside the project team type the five or six things a neighbor would type into ChatGPT, Gemini and Perplexity, and save every answer with the date. Read them the way the daughter in the city would. Then publish the plain answer each one was missing, on a page that carries your company’s legal name, the project name and the docket number, so the model has no reason to confuse you with anyone else. If an old incident shows up in the answers, deal with it on that page in plain terms: what happened, what you changed, and what the regulator concluded. Silence doesn’t make the old story go away. It leaves the old story as the only source.
Give this job to whoever owns community relations, not to marketing. The questions come from neighbors, and the answers have to hold up when a landowner’s attorney reads them next to your filings. If the plain-language page and the docket disagree, the disagreement becomes the story.
At ModalPoint we measure what AI assistants say about energy companies, so far mostly the answers that buyers and job candidates get. The community side is the same problem with different questions. You don’t need us to run the first test, though. It takes an afternoon.
Start with what people have already said on the record. For an interstate gas project, that is your docket in FERC’s eLibrary, where the public comments are filed. For a compressor station going through a township or county zoning hearing, it is the minutes and public comment from the last few sessions. Either way, notice which questions keep coming back. Those are the same questions people are typing into a chatbot, usually with less patience.
What I don’t know yet is whether a chatbot answer has changed the outcome of a permit fight anywhere. I suspect it already has, quietly, through who showed up and what they believed when they got there, but I haven’t seen a case documented. If you’ve watched an AI answer get quoted at a public meeting on one of your projects, I’d like to hear what it said and how your team handled it.
Matthew Bertram is Chief Marketing Officer of the Oil & Gas Global Network, President of ModalPoint, an AI decision-governance advisory, and CEO of EWR Digital, a Houston digital marketing agency in business for more than 25 years. He hosts The Best SEO Podcast, created the LLM Visibility™ methodology for getting brands cited in AI search, and is a contributor to NIST’s Cyber-AI Profile.