01 · The shift Where your brand is described now.
Brand visibility used to mean a position in a list of links. A person read your own words on your own page. That is no longer the first contact.
Today a supporter asks an AI model a question. The AI model writes one answer. It builds that answer from pages it can reach, read and quote. Your brand appears inside somebody else's sentence, or it does not appear.
Your website still carries this. Among donors who research a charity, 54 percent read the charity website. Only 7 percent read an AI summary. The website is both the destination and the source, so a site that AI models cannot read weakens both.
02 · The cost What a shortfall in awareness costs you.
An AI model that cannot reach your pages still gives an answer. It writes that answer from a directory listing, an old news article, or a peer organization's page.
Three things follow, and none of them show up in your analytics:
- Somebody else gets the introduction. The supporter meets a version of your mission that you did not write.
- Your own numbers go missing. Your impact figures, your programs and your donation page never enter the conversation.
- Old facts harden. A stale address, an old leader name or a closed program keeps getting repeated as current.
This is a fundraising problem, not a technical one. It is also usually cheap to correct. In most first audits, a small number of faults explains most of the loss. One of them is often a single line in a file that a web vendor set years ago.
Do we block OAI-SearchBot, PerplexityBot or Claude-SearchBot in robots.txt? A "yes" cancels every other investment in brand visibility.
03 · The report What the audit gives you.
We built a measuring instrument, not an opinion piece. It answers one question. Can AI models find your organization, read it and quote it correctly?
- A score out of 100, and a repair list in date order. The workbook does the scoring for you.
- An owner and a time estimate for each section. You can hand the work to three people and finish faster.
- 20 test prompts. The exact questions to ask ChatGPT, Claude, Perplexity and Gemini, with the method that makes the answer mean something.
- A list of eight recommendations that no primary source supports. Several appear in paid proposals. Read that section before you approve a budget line.
Every claim in the report cites a primary source. We state the year, the sample size and any commercial interest behind each one. We also publish what the evidence does not support, because a document that only lists what to do is a sales document.
A failed check in Section A cancels your total. An organization with 92 points and a blocked search bot does not appear. Start with Section A. It takes about 30 minutes and it decides everything after it.
Most small nonprofit sites score between 45 and 70 on a first pass. A low first score is normal. It measures the site, not the team.
04 · Objections The questions nonprofit teams ask us.
These come up in almost every first conversation. The answers below are short. The report gives the source for each one.
"We have a small team and no developer. Can we still do this?"
Yes. The audit needs no code. Each of the six sections names an owner and a time estimate, so a communications lead and a finance lead can take a section each. One person finishes in about 90 minutes.
"We already pay an agency for SEO. Is this different?"
Yes. Search engine optimization aims at a position in a list of links. This measures whether AI models can reach your pages, read them, and quote them correctly. Ask your agency to name the primary source for each deliverable they propose. Section 3 of the report tells you which answers to reject.
"Can you guarantee that AI models will recommend us?"
No, and nobody can. AI models give different answers to the same question, and no vendor publishes its ranking method. Decline any proposal that promises you a citation. A citation is also not a recommendation.
"We just rebuilt our website. Are we covered?"
Not always. A new site can score worse than an old one. Most AI models do not run JavaScript, so a site that builds its content in the browser sends them an empty page. Section B measures this in about 10 minutes.
"Most of our program information sits in PDFs. Does that matter?"
Yes. Move the information that supporters ask about to an HTML page first. Keep the PDF for people who want to print it.
"Only 7 percent of donors read an AI summary. Is this worth our time?"
Your website feeds both channels. The same faults that hide you from AI models also hurt the 54 percent who read your website. You repair one thing and both improve.
Sources and method: This is an introduction to The AI Visibility Audit, Socient Research Lab report SRL-2026-01, published 15 August 2026. The report reads each primary document and states the year, the sample size and any commercial interest behind every source. Crawler behavior and access controls come from published vendor documentation from OpenAI, Anthropic, Perplexity and Google Search Central. JavaScript execution by AI crawlers comes from a measurement of about 1.3 billion crawler requests. Donor research figures come from the BBB Wise Giving Alliance Donor Trust Report 2026. Section 3 of the report names eight recommendations that no primary source supports, and gives the refuting source for each one. Analysis and recommendations are Socient's own.