Open-source AI-marketing agents: a free stack to find where AI search ignores you

Foundations

Why is a mention not the same as a citation?

AI search is a new front door, and it is opinionated. Ask an engine “what’s the best payment stack for SaaS founders,” and it returns a few named sources — not ten blue links. If your company is not one of them, you are invisible at the moment of intent. No analytics tab tells you so.

A brand mention is when the model names you in its prose. A citation is when it attributes the answer to a URL on a domain you own. These are not the same thing. In our own audits we have watched an engine praise a company in one sentence, then source only its competitors’ pages. Mentions are nice. Citations compound, because they send the engine — and the reader — back to a property you control.

We build AI-marketing agents at Humanswith.ai, and we run this loop — measure, produce, optimize, design — for clients every week. In 2026 I open-sourced the core of how we work: four free tools, no sales call, that are the transparent skeletons of the agents inside our hosted Humanswith.ai Workspace. The hosted version automates the scale and the weekly rhythm. The free tools give you the method to start.

Measure

How do you measure where AI search ignores you?

You cannot improve what you have not measured, and the first measurement is simple: ask the engines and read the answers. The open-source ai-visibility-probe-lite is a small kit for exactly that. You write brand-free discovery prompts — the questions a buyer types before they have heard of you (“best X for Y,” not “is Acme any good?”), run them yourself in ChatGPT, Perplexity, or Gemini, and paste the answers back in.

The probe then computes two distinct numbers: mention-share (how often the answer named you) and citation-share (how often it cited a URL you own), for you and the competitors you list. It also runs a transparent Mini Briefing Test that scores whether a page is even citable. Brand-free prompts keep you honest: asking “is my brand good?” measures sentiment about you, not whether the engine treats you as an authority for the category.¹

Produce + Publish

How do you produce content worth citing?

The fastest way to write uncitable content is to paste a topic into a chatbot and ship the first draft. The open-source contentos-agent-lite does the opposite: a content agent you run inside your own coding assistant that writes from a documented process, not a prompt. It walks eight gates — business context, research, a source pack, a brief, a draft, an editorial pass, a publish-readiness check, and distribution — and it refuses to invent a fact that has no source.

Its eighth gate is the one most teams skip: gate 08 is canonical-first distribution. Before you publish, it lints the page itself — one canonical URL, one H1, Open Graph tags, Article structured data — the publishing-side signals that decide whether an engine can fetch and attribute a citation to your page. Then it drafts adaptations for LinkedIn, Medium, and dev.to, and every draft links back to your canonical URL. Publish on your own domain first, and make every copy on a platform you rent point home, so the authority compounds where you own it.²

Optimize

Why can a well-written page still go uncited?

A page can be perfectly written and still never get cited, because three gates sit in front of every citation. The open-source aeo-site-audit-lite checks all three: Fetchable (can a crawler reach and index it — HTTP status, robots/noindex, canonical, structured data), Chosen (among the pages it can fetch, why does it cite the other one — you supply the URLs an engine cites today and the tool classifies the gap: authority, freshness, a competitor’s data table, missing proof), and Extractable (once chosen, can it lift a clean answer out — headings, a concise lead answer, schema, list and table readiness).

Point it at a URL or a local file and you get a scored report and a prioritized fix list. It runs offline, needs no keys, and uses only the Python standard library — so it is safe to read before you trust it.³

Design

How do you keep distributed content on brand?

Citable content still has to look like you when it surfaces in a feed. The open-source brand-card-lite turns a tiny brand-tokens file — your colors, fonts, and logo — into a self-contained, on-brand social card, and lints those tokens for contrast and consistency: text-on-background that actually meets WCAG, fonts with real fallbacks. No image model, no cloud service, no keys — templating and a little math you can audit.⁴

Pitfalls

Where do companies go wrong with AI visibility?

Four mistakes show up again and again, and each tool above targets one of them.

Playbook

How do you run the loop this week?

You do not need the hosted product to begin. Here is the smallest honest version of the loop:

1. Measure. Run ten brand-free prompts through one engine with the probe. Record mention-share and citation-share.
2. Pick one loss. Choose a prompt where a competitor is cited and you are not.
3. Write the answer. Take that prompt through the content agent’s gates, then publish it on your own domain first.
4. Audit the page. Run the site auditor and close the top Fetchable, Chosen, or Extractable gap.
5. Dress it. Generate an on-brand card so the post looks like you in the feed.
6. Re-measure next week. Run the same prompts again and watch the two numbers move.

The bet

Why give the core away for free?

These are honest skeletons, not crippled trials. The method — brand-free measurement, the mention-versus-citation split, the eight content gates, the three retrieval gates, canonical-first distribution — is the part that changes outcomes, and it is fully in the open. The hosted Humanswith.ai Workspace adds the part you cannot do by hand at scale: automated multi-engine scans on a weekly cadence, the publishing and re-measurement loop that proves whether a fix worked, and the team and hosting around it. The free tools tell you where you stand. The workspace runs the loop for you.⁵

Open-sourcing the core is a deliberate bet: the companies that win AI search will treat it as an operating loop, not a one-off audit. The fastest way to start is to clone a tool, measure one prompt set, and fix one page this week.

Sources

References and source notes

FAQ

Frequently asked questions

Related pages

Where this page connects inside the site