AI Visibility for Nonprofits: Donor Prompts, Stale Facts and Fixes

People rarely type your name first. They describe a cause, a place or a worry, and the assistant picks the organizations. Test prompts that match each audience:

• "What are the most effective charities fighting childhood hunger in Chicago?"
• "Is [organization name] a legitimate charity, and are donations tax-deductible?"
• "Where can I volunteer on weekends with animals near Denver?"
• "Who runs [organization name] and how much goes to programs versus overhead?"
• "Which nonprofits accept donated laptops for students?"
• "Is [organization name] the same as [similarly named group]?"
• "What foundations fund literacy programs for adults in rural Ohio?"

The first and fifth prompts decide whether you appear at all. The second, fourth and sixth decide whether the assistant gets you right once you do. Run both kinds. The free checker on this site runs up to three buyer questions and shows whether the engines name you, which is a fast baseline before you build a fuller prompt list.

Donors use ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Claude and Microsoft Copilot. Each pulls from a different place. Google's AI features draw on Google's index. Copilot grounds on Bing. ChatGPT, Perplexity and Claude run their own crawlers and searches.

Stale answers usually come from the parametric stage: what the model learned in training, with no live search. OpenAI's model docs (checked August 2026) list a February 16, 2026 knowledge cutoff for the GPT-5.6 family and August 31, 2025 for the faster Instant path. If your executive director changed in spring 2026, an answer given without search can still name the old one. Retrieval is per query and at the model's discretion, so the same prompt can return a fresh answer one day and a stale one the next.

What it means: you can't fix the frozen layer this quarter. You can make sure the live layer finds current facts. Compare both with the search on vs off test, and read how a new model release can change what it says about you.

Nonprofits change slowly in public and quickly in reality. These are the facts to check in every answer:

• Leadership. Former executive directors, board chairs or founders named as current.
• Mission. An old mission statement, or a mission blended from two organizations.
• Programs. Ended programs described as active, new ones missing.
• Tax status. The wrong exemption category, a missing registration number, or a claim that donations are not deductible.
• Service area. A local chapter described as national, or the reverse.
• How to give. Defunct donation links, a closed thrift store, a wrong mailing address.
• Name conflation. Two groups with similar names merged into one answer.

Say you run a hypothetical food bank called Riverside Community Pantry. An assistant might attach the programs and leadership of a Riverside Community Pantry two states away, or say you were founded the year the other group was. That mistake can send a donor's gift to the wrong organization. Record each error with the engine, the date and the exact wording, so you can see whether it changes.

When engines cite other sites instead of yours, being named correctly on those sites is the route into the answer. That is an association in the evidence, not a promised effect. For nonprofits, the sites that matter are:

• Charity evaluator profiles. Check name, mission, leadership, financial year and ratings. Use each evaluator's correction process.
• Government charity registers. Your national or state registry is the record of your legal name, registration number and status. Make sure your filings match what your site says.
• Your Wikidata item. Wikidata is a free, collaboratively edited knowledge base. Check that your item exists, points to your official website, has the right founding date and location, and isn't merged with a similarly named group. Edit with references and follow Wikidata's norms, because promotional edits get reverted.
• Volunteer platforms, grant databases and local directories. These often carry years-old descriptions. See whether being listed in directories helps AI answers.
• Community discussion. Forums and threads shape how people describe you. Read about how Reddit discussion relates to AI answers.

Fix errors at the source. Correcting your own website does nothing for a wrong evaluator profile that an engine keeps citing.

A perfect fact page does nothing if AI crawlers can't fetch it. Check these in order.

robots.txt, bot by bot

Rules are per bot. OpenAI states each setting is independent: GPTBot covers training, while OAI-SearchBot controls whether you appear in ChatGPT search. A blanket AI block often removes search eligibility by accident. Check OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and bingbot separately.

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

That pattern stays in ChatGPT search while opting out of training.

Firewall and CDN

A firewall or CDN can block a crawler that robots.txt allows, often with a 403. Many nonprofits run on free CDN tiers. From September 15, 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, while Search stays allowed. Check your settings rather than assume. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.

JavaScript and snippet controls

Vercel found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. If your mission, programs or donation details load only through a script or widget, those crawlers may see a blank page. Google renders JavaScript, but it needs the page indexed and snippet-eligible for AI Overviews. Look for stray noindex, nosnippet, max-snippet:0 or data-nosnippet from a template.

The one positive experimental signal for on-page content is simple: a specific fact stated plainly on a fetchable page gives the engine a sentence it can quote. For nonprofits, put these in the server-rendered HTML of your About page, in plain sentences:

• Legal name, any former names, and how you differ from similarly named groups.
• Tax status and registration number, and whether gifts are deductible where you operate.
• Current executive director and board chair, with the year they started.
• Founding year, headquarters and service area.
• Active programs, and programs you have ended.
• How to donate and volunteer, with working links.

Use the same wording on your site, your filings, evaluator profiles and Wikidata. Contradictions within your own site give engines two versions to choose from. For what this means on one engine, see getting cited by ChatGPT.

1. Days 1 to 5: baseline. Run 5 to 8 prompts across ChatGPT, Google AI Overviews, Gemini, Perplexity, Claude and Copilot. Log every wrong fact with engine, date and wording.
2. Days 6 to 10: access. Read robots.txt per bot. Test each crawler against your firewall or CDN. Compare raw HTML with the rendered page on About, Programs and Donate.
3. Days 11 to 18: your own site. Fix JavaScript-only content and stray snippet controls. Publish the plain fact sentences above, and remove contradictions between pages.
4. Days 19 to 26: third parties. Submit corrections to evaluators and registers. Review and reference your Wikidata item. Update volunteer and grant listings.
5. Days 27 to 30: re-check. Re-run the same prompts. Expect live-search answers to move before training-based ones, and record what didn't change.

None of this guarantees a mention or a citation. It removes the blocks you control and gives engines correct facts to find.

Does allowing GPTBot help ChatGPT recommend our charity? No. OpenAI documents GPTBot as a training crawler only. ChatGPT search uses OAI-SearchBot, and the two settings are independent. You can block GPTBot to opt out of training and still allow OAI-SearchBot to stay eligible for ChatGPT search. Nothing in OpenAI's documentation says allowing GPTBot improves search visibility, so decide on training use as a separate policy question.

Should we add Organization or FAQ schema to get cited? Schema is not a lever for AI answers. Google says structured data isn't required for generative AI search. Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema, though those pages were already heavily cited. Existing markup does no harm, so there's no need to remove it. Spend the time on plain, fetchable fact sentences and access checks instead.

Why does ChatGPT still name our former executive director? Most likely the answer came from the model's training data rather than a live search. Models carry a knowledge cutoff, and answers given without retrieval reflect what was true then. Make sure the current leader is named in plain HTML on your site and on evaluator profiles, so searches find it. Then re-test with search on and off to see which layer is still stale.

Another group has almost the same name. What can we do? State the difference in one plain sentence on your About page: your legal name, registration number, location and founding year. Make sure your Wikidata item and evaluator profiles carry the same identifiers and point to your official website. Check that the other group's Wikidata item isn't merged with yours. You can't stop a model from confusing you, but you can make the distinction easy to find.