Spreadsheet-supported review
The spreadsheet version keeps the human judgement and removes the arithmetic errors. The registry sits in one sheet with a stable question identifier. Coding fields use dropdowns rather than free text, so "yes" always looks like "yes". Formulas compute mention rate and share from the coded rows instead of from someone's recollection.
This suits a team of two reviewers who handle 50 to 150 questions across two or three engines and double-code roughly a fifth of rows to check agreement. It's where most nonprofit competitive intelligence should live for the first year, because the cost is staff time you already spend and the method is fully visible to whoever asks.
Add a notes column for anything the coding fields can't capture, such as a dated claim or a confusion with a similarly named charity. Those notes become the evidence behind your corrections. A disciplined AI SERP watch run this way beats an expensive nonprofit competitive intelligence tool that nobody opens.
Specialist monitoring platforms
Platforms automate the collection and repeat the sample daily. Published entry pricing in 2026 runs from $29 a month for 15 prompts on Otterly.ai's Lite tier through $95 a month for 50 prompts on Peec AI's Starter plan, with extra engines charged as add-ons. Scale comes from prompts multiplied by models multiplied by frequency, so the bill grows faster than the registry does.
Verify four things before you report a platform's score to anyone. Ask for the written denominator, because an audit of nine platforms on 10 September 2026 found three vendors publishing contradictory definitions of their own visibility metric, where a score of 50 can mean half of all answers or half of the answers that named any organisation. Ask whether raw answers are retained and exportable. Ask how many runs sit behind a daily figure. And ask whether the competitor set is one you define, since a closed list you built yourself improves the moment you remove someone from it.
Platform visibility share tracking is useful when the registry is already stable and the reporting burden is the bottleneck. It's a poor first purchase. Specialist monitoring platforms can help once the underlying method for nonprofit competitive intelligence is sound.
Turn findings into actions
Evidence without an owner decays into a slide. Build a decision log with one row per finding, and close each row with a dated follow-up check rather than an assumption that the fix landed.
-
The saved evidence, with engine, date and question identifier
-
What the finding is, in one sentence a trustee would understand
-
Named owner and the priority you've assigned
-
The action taken and the date of the re-check
Sort findings by type. A factual correction, such as a wrong service area or a programme you closed in 2023, goes straight to whoever controls your website and your Candid profile, which feeds the data behind more than 200 philanthropic websites and applications. Missing methodology is a different job: if answers describe your impact claims vaguely, the fix is publishing the method behind the number. Page improvements belong with content staff, and source context belongs with whoever manages media relationships.
Technical investigation deserves its own lane. Work out whether the pages you expect to be cited are reachable and current before you conclude that a model dislikes you. And separate material errors from routine variation, because an organisation dropping out of two answers one week isn't evidence of anything. Nonprofit competitive intelligence earns its keep when it changes a page.
Set a defensible cadence
Run a baseline across the full registry, with repeats, and treat it as the reference everything else compares against. After that, re-run on a quarterly schedule and additionally after a reorganisation or renaming on your side. Re-run after a substantial content release and after a known platform update.
Keep the questions stable between runs. Document additions with the date they entered the registry and report them separately until they have a baseline of their own. Platform behaviour shifts underneath you regardless: Semrush's three-month study tracked Wikipedia falling from appearing in roughly 55% of ChatGPT responses to under 20% in September 2025, while its share on AI Mode and Perplexity held steady. A drop in your own numbers across that window would have told you almost nothing about your content.
Say so in the report. An AI SERP watch in nonprofit competitive intelligence describes a sample of answers on named engines in a named week, and visibility share tracking inherits every limitation of that sample. Writing the caveat into the template protects the credibility of the next finding, the one where you're right and something genuinely needs fixing.
The scale behind all this is worth holding in mind. ChatGPT reached 800 million weekly active users by October 2025, and Blue State's November 2025 analysis cited by Candid found nonprofit organic search traffic down 35% in a year while AI-referred traffic rose more than 1,000%. Klaudia Jaźwińska and Aisvarya Chandrasekar, the Tow Center researchers behind the citation study, put the mechanism plainly: "generative search tools parse and repackage information themselves, cutting off traffic flow to original sources."
Pilot nonprofit competitive intelligence
Pick one decision. Build 25 questions around it and run them twice. Code the answers and see whether the evidence changes what you do. Then ask whether a wider registry would have changed anything else. Most organisations find the first pilot pays for itself in corrections.
Snoika Foundation is a Tallinn-based team that helps nonprofits and public institutions become visible and correctly cited in ChatGPT and Google. We run free AI visibility audits for mission organisations. Book a call with us and we'll scope a pilot in nonprofit competitive intelligence around the decision you're facing now.