AI Search SEO Agent (AISO)

The next layer of SEO: measures whether AI answer engines (Perplexity, ChatGPT, Gemini, Claude) mention and cite your brand — with real engine probes, never invented scores — then closes the gaps with entity fixes and citation-friendly content auto-published via GitHub PR.

View in marketplace Open dashboard

The AI Search SEO Agent (agent #18, Marketing) optimizes for AI Search Optimization (AISO): being recommended and cited inside AI-generated answers, not just ranked on Google. Each assessment it asks your seed questions to every configured answer engine and records, per query × engine, whether your brand appeared and whether your domain was cited (with the actual source URLs). From those real probes it computes three 0–100 scores — AI Discoverability, Citation Probability, and Entity Authority — audits your Schema.org/Wikidata/Wikipedia entity footprint, runs a competitor gap analysis on who the AIs recommend instead of you, and turns the gaps into citation-friendly articles with FAQ and JSON-LD that it can push to your GitHub repo as a PR-ready branch. Claude works out of the box on the platform key; add PERPLEXITY_API_KEY / OPENAI_API_KEY / GEMINI_API_KEY for full multi-engine coverage — engines without a key are simply skipped. White-hat only.

What it does

This agent answers the question classic SEO can't: when someone asks ChatGPT or Perplexity "best X for Y", are YOU the answer? Its core loop is measurement-first and grounded by an absolute rule — every visibility number must come from a real engine response. aiso_simulate runs up to 12 queries across the configured engines (Perplexity's sonar with its real citations array; ChatGPT via gpt-4o-mini and Gemini via gemini-1.5-flash with URL extraction and mention detection; Claude on the platform key), detecting for each probe whether the brand appeared in the answer, roughly how early, and whether the domain was among the cited sources. Probes are persisted per run, and the scores are computed directly from them: Discoverability = the percentage of probes where you appeared; Citation = the percentage where you were cited. aiso_entity_audit checks the machine-trust signals LLMs lean on — Schema.org JSON-LD on your homepage (40 points), a Wikidata entity (30), and a Wikipedia page (30) — returning Entity Authority plus the specific gaps. aiso_competitor_gap re-runs your queries and reports, per query, which named competitors the engines mention when you're missing.

Then it acts on the gaps. aiso_optimize_content generates AISO-shaped content: a one-paragraph direct answer up top, clear H2/H3 structure, fact-dense paragraphs that make good retrieval chunks, a key-facts list, an FAQ, and ready-to-embed Article + FAQPage JSON-LD. With GitHub connected and Autonomous mode plus a repo configured, aiso_publish_content ships it: a new aiso/<slug>-<timestamp> branch off your default branch with the content committed and a link to open the Pull Request — never a commit straight to main; in Recommend mode content stays as drafts for you. Every run finishes with a saved report (scores, per-engine findings, entity and competitor gaps, prioritized fixes with the WHY) in the AI Search console, and usage is metered per engine probe. Run it on a schedule to track visibility movement over time. Everything is white-hat: accurate, genuinely useful content and correct structured data — no cloaking, fake entities, or spam.

How it works

Your Business

Simulate real user questions

aiso_simulate asks each seed question (informational / commercial / navigational — or ~8 generated ones if you set none) to every configured engine: Perplexity (sonar, real citations), ChatGPT (gpt-4o-mini), Gemini (gemini-1.5-flash), and Claude (platform key). Each probe records appeared, cited, approximate position, a snippet, and the cited source URLs. Engines without an API key are skipped.

Score from the probes

The run persists every probe and computes the scores directly from them: AI Discoverability = % of probes where the brand appeared; Citation Probability = % where the domain/brand was among the cited sources; plus a per-engine appearance breakdown. aiso_score returns the latest KPIs any time.

Audit the entity footprint

aiso_entity_audit fetches your homepage for Schema.org JSON-LD (40 pts), searches Wikidata for your entity (30 pts), and checks for a Wikipedia page (30 pts) — producing the 0–100 Entity Authority score and the concrete gaps suppressing AI trust.

Find the competitor gaps

aiso_competitor_gap re-runs your queries (up to 8, vs up to 5 competitors) across the engines and reports per query whether you appeared and which competitors were mentioned instead — your prioritized content/entity hit-list.

Generate citation-friendly content

aiso_optimize_content writes an AISO-optimized article: direct-answer intro, clean H2/H3 structure, fact-dense retrieval-friendly paragraphs, a key-facts list, an FAQ, and ready-to-embed Article + FAQPage JSON-LD. White-hat: accurate and genuinely useful, no keyword stuffing.

Publish via GitHub PR (optional)

With GitHub connected and a repo configured, aiso_publish_content commits the content to a new aiso/<slug>-<timestamp> branch off your default branch and returns the link to open the Pull Request — never straight to main. Autonomous mode publishes; Recommend mode leaves drafts.

Report and repeat

aiso_make_report saves the full picture — scores, per-engine findings, entity + competitor gaps, prioritized recommendations with WHY — to the console's Reports tab, then record_output closes the run. Scheduled re-runs build the visibility-over-time trend.

Outcomes delivered

Setting it up — owner / admin

1
Subscribe and open the AI Search console
Subscribe from the marketplace, then work from the console at /dashboard/aiso — score KPIs, the Runs list (per-probe detail), Reports, and the Run assessment button live there; configuration is on the deployment page at /dashboard/agents/[id].
2
Set the brand, domain, and seed questions
Fill Brand / entity name and Website domain, then the Seed questions you want to win in AI answers (one per line — e.g. 'best lost-pet finder app'). Leave questions empty and the agent generates ~8 realistic ones across informational/commercial/navigational intent.
3
Add engine keys for full coverage
Claude works out of the box on the platform key. Add PERPLEXITY_API_KEY, OPENAI_API_KEY, and GEMINI_API_KEY to cover Perplexity, ChatGPT, and Gemini — any engine without a key is skipped, and scores note which engines ran.
4
List competitors
Add the brands the AIs might recommend instead of you (one per line) so the competitor gap analysis can show exactly which queries they win.
5
Choose autonomy and (optionally) a publishing repo
Recommend mode measures and drafts content for you to publish; Autonomous mode also opens content PRs. For publishing, connect GitHub under /dashboard/tools and set the repo (owner/name) in config.
6
Run the first assessment, then schedule
Hit Run assessment in the console and watch the run land with per-engine probes. Then set Run frequency (Daily / Weekly / Monthly) so re-runs track your AI visibility over time and catch drops.

Using it day to day — your team

1
Read the three scores
The console shows the latest AI Discoverability, Citation, and Entity Authority (each 0–100) plus queries run — all computed from that run's real engine responses.
2
Drill into a run
Open any run to see every probe: the query, the engine, whether you appeared and were cited, the answer snippet, and links to the cited sources — exactly what the AI said.
3
Watch the trends
The visibility-over-time chart tracks Discoverability and Citation across runs, and the per-engine bar chart shows where you're strong (e.g. cited on Perplexity) vs invisible.
4
Work the prioritized fixes
Each report lists entity gaps (e.g. missing JSON-LD, no Wikidata item) and the competitor-won queries, with the WHY behind each recommendation — fix in priority order.
5
Review and merge content PRs
In Autonomous mode with a repo set, optimized articles arrive as PR-ready branches on your repo — review the content and JSON-LD, open the PR, and merge to ship. You can also chat with the agent; chat gets read-only tools (score, entity audit, competitor gap).

Use cases

Am I in the answer?

Simulate 'best X for Y' questions across Perplexity, ChatGPT, Gemini, and Claude and see — per query, per engine — whether you appear, whether you're cited, and what sources the AI actually used.

Close the competitor gap

Find the exact queries where the AIs recommend competitors and you're missing, then generate the targeted citation-friendly content and entity fixes to win them.

Build entity authority

Audit and fix the machine-trust signals LLMs rely on — Schema.org JSON-LD on your site, a Wikidata entity, Wikipedia presence — with a concrete 0–100 score and gap list.

Monitor AI visibility over time

Scheduled assessments re-probe the engines so you see Discoverability and Citation trends, catch drops early, and prove the impact of published fixes.

What to expect

An AI visibility report with three grounded 0–100 scores — Discoverability, Citation Probability, Entity Authority — plus a per-engine breakdown
Prompt-simulation evidence: per query × engine, whether you appeared and were cited, with snippets and the actual source URLs
An entity audit (Schema.org / Wikidata / Wikipedia) with the specific gaps suppressing AI trust in your brand
A competitor gap map showing which queries the AIs answer with rivals instead of you
AISO-optimized articles with FAQ + Article/FAQPage JSON-LD, delivered as drafts or auto-published to a PR-ready GitHub branch — never straight to main

Metrics to watch

AI Discoverability — % of probes where the brand appeared; the headline number to move
Citation Probability — % of probes where your domain was among the cited sources (mentions without citations are the next gap)
Entity Authority — the schema/Wikidata/Wikipedia composite; usually the fastest score to fix
Per-engine appearance rate — which engines know you and which don't (coverage depends on which API keys are set)
Competitor-won queries — how many of your target queries still surface rivals instead of you, run over run
Published-content impact — Discoverability/Citation movement on targeted queries after PRs merge, plus metered probe usage per run

FAQ

How is this different from the SEO Agent?

The SEO Agent optimizes classic search rankings (Google SERPs, crawls, Search Console). This one optimizes AI answer engines — whether ChatGPT, Perplexity, Gemini, and Claude mention and cite you inside their answers. They're complementary layers; run both.

Are the scores real or estimated?

Real, by hard rule: the agent may not invent a visibility number. Every score is computed from that run's persisted engine probes — Discoverability is literally the % of probes where you appeared, and Citation the % where your domain was in the cited sources. Open any run to see the underlying answers and sources.

Which engines are covered, and do I need API keys?

Perplexity, ChatGPT (OpenAI), Gemini, and Claude. Claude works out of the box on the platform key; add PERPLEXITY_API_KEY / OPENAI_API_KEY / GEMINI_API_KEY for the rest. Engines without a key are simply skipped — you still get scores from whatever ran.

Can it publish content for me, and is that safe?

Yes — with GitHub connected under Tools, a repo configured, and Autonomous mode on, it commits each optimized article (with its JSON-LD) to a fresh aiso/ branch off your default branch and gives you the link to open the PR. It never commits straight to main, and in Recommend mode nothing is published — content stays as drafts.

Is this AI-answer manipulation / black-hat?

No. The system enforces white-hat only: accurate, well-structured, genuinely useful content and correct structured data (Schema.org, Wikidata). No cloaking, fake entities, keyword stuffing, or spam — those patterns get brands demoted by engines, not cited.