Which AI-Visibility Tools Have an API, and What Plan Carries It?

API availability and API entitlement are different buying questions. This report originally put numbers on both for the 27 August 2026 CitedIndex corpus. The population can be recovered; the published API and plan-placement totals cannot be reproduced reliably from the evidence recovered for this correction. They are withdrawn as research findings, not silently replaced with today’s counts.

Editorial correction, 5 September 2026: corrected the unsupported historical totals, the inference that missing plan text means sales-only access, and the description of self-hosted API access as cost-free or necessarily open source. The original publication and verification dates remain. This was a read-only archive review and a limited check of the linked public documentation, not a new census, account-level access test or pricing sweep.

Corpus and method

102 Published listings were captured in CitedIndex warehouse collector run 221171.

86 of those captured listings remain after excluding the 16 carrying the citation-services category in that same run.

The capture is labelled 27 August 2026 in Europe/Amsterdam and was observed at 2026-08-26 22:00:03 UTC. These are directory membership counts, not an estimate of the whole market. Agencies and other managed services in the excluded category were outside the software comparison; exclusion does not mean that an agency cannot provide an integration. Later additions and category changes are not backdated into this population.

The original method counted the API-availability field, then searched the recorded plan feature text for an included API. It excluded explicit denials, customer-supplied model keys and framework names such as FastAPI. That remains a useful screening method, but a text match alone cannot establish endpoint scope, entitlement or a fully loaded price.

Limitations and the historical correction

The archived listing-content versions overlapping the capture do not agree with the article’s availability split. The claim registry also retained an earlier edition’s figures and queried mutable live rows, rather than the population used by the article. Neither a revision note saying a recount happened nor a count of today’s listings supplies the missing historical proof. We have not established which original classifications caused the disagreement, so this correction does not publish a new API-availability total.

The following is a record of the withdrawn 27 August claims, not a replacement census or a set of figures to cite:

Original claim	Correction status
76 determined software tools: 53 API yes, 23 no, 10 undetermined	Withdrawn as a verified availability split; the preserved population is 86 software listings
36 of the 53 place API in a named plan; 17 do not	Withdrawn pending a reproducible row-level entitlement classification
11 unplaced with plan text and 6 without feature-level plan text	Withdrawn; missing research text is not evidence of vendor silence or sales-only access
8 quote-only, 28 with a published-price plan, 3 sold both ways	Withdrawn as category totals; a published price also need not include usage or onboarding charges
5 zero-dollar API plans; 23 paid, split into 19 dollar and 4 euro prices	Withdrawn as a verified market count and price distribution
$14.99–$999 and €79–€299 monthly API entry-price ranges	Withdrawn as census findings; historical quotes are not current purchase offers

1. A product API is not a model-provider key or an MCP server

A conventional REST API exposed by the product can let a customer’s backend retrieve data or start work. A key supplied to OpenAI, Gemini or a scraping provider instead pays for a dependency used by the product. One does not prove the other. An MCP server is also a separate interface: the MCP architecture defines a host/client/server protocol, not a promise that the vendor exposes a REST endpoint.

Canonry’s technical surface explicitly documents a REST API and an OpenAPI route, alongside a separate MCP adapter. That is stronger API evidence than a framework name. Syntropic AI Visibility Audit’s README distinguishes its self-hosted application from the external provider keys it requires. Its FastAPI backend label alone does not establish a supported customer-facing integration contract. The earlier assertion that its API was unquestionably real despite only a stack description in our plan text was too strong.

Method: compare the documented interface and prerequisites, rather than count the word API. No authenticated API call was made for this review.

2. Missing plan text does not mean an API requires a phone call

The original unplaced list was AEOlens, Conductor, LLM Pulse, Mangools AI Search Watcher, Mentionable, Morningscore, Promptwatch, Rankability, Sleepwalker, Syntropic AI Visibility Audit and Writesonic. It was a list of what our stored plan text failed to place, not proof that those vendors withhold entitlement or require a negotiated price. The companion research-gap list was CiteMentor, Distribb, Elmo, Indexly, Keyword.com AI Visibility and Orchly AI. Those omissions described our record, not the completeness of the vendors’ own pages. Neither list is a current classification.

A concrete correction: AEOlens’ pricing page now explicitly names API access in its Enterprise offer. That supports an Enterprise placement in the page read for this correction; it does not establish when that wording appeared or prove that no other access arrangement exists. We withdraw the categorical statement that AEOlens does not attach its API to a plan.

Rank Prompt’s MCP documentation is more explicit about the shared entitlement: Starter and higher, or a standalone API plan, cover its MCP server and public /v1 API. That is a vendor-specific statement, not a rule to apply to every product advertising both interfaces.

Method: read a named entitlement statement. Missing rows in our research stay unknown rather than becoming a vendor negative.

3. Keep quote-only, read-only and self-hosted separate

The original quote-only list named Cognizo, Dageno AI, Emergine, GeoArk AI, GeoRankers, Goodie AI, RadarKit and Scrunch AI. The list remains here to make the correction auditable, but its exclusivity and completeness are withdrawn. In particular, GeoRankers’ pricing places “Agent-Ready API (MCP-compatible)” on Enterprise. That supports the advertised Enterprise feature; it does not by itself prove a conventional REST API or an independently connectable MCP server. Those require the actual interface documentation.

The original free-plan examples were AI Sightline, Canonry, GEOrank, geoSurge and RankLens. They must not be read as five integrations with no operating cost. Self-hosting, model calls and a bring-your-own-key dependency can cost money even when the application has no subscription charge. Canonry’s license is FSL-1.1-ALv2, restricting competing use before its future Apache license takes effect; calling that release simply “open source” omitted a material restriction.

Read-only access can be sufficient for pulling a report. It is a limitation if the task is to create prompts or trigger a run, not a universal defect. The same applies to an Enterprise-only API: it can exist without being available on the plan a buyer intends to purchase.

Method: separate interface existence, allowed operations, plan entitlement, billing interval and external costs. The retired low/high examples were TrueRanker and ApexGEO in dollars, and Omnia and Surfer SEO in euros; this correction neither repeats them as offers nor converts currencies.

Buying checklist

Before integrating, obtain the documented endpoint, authentication method, allowed data and operations, named plan and any add-on requirements. Check pagination, export history, rate and usage limits, and whether running a report consumes credits beyond retrieving it. Ask for a sample response generated under the plan you would buy; do not infer entitlement from an in-product dashboard or from a vendor’s own use of a model API.

For the separate external-agent question, see the MCP server report. The dated engine-coverage report explains why engine breadth also does not establish plan-level access.

Citing this correction

Cite the recovered directory population and the linked interface distinctions, not the withdrawn availability, plan-placement or price totals. A complete corrected historical API census would need the original per-listing classifications and their contemporaneous source support; this page does not pretend that a current documentation check recovered them.