Visibilità AI nelle query di confronto con competitor

FirstPosition.ai measures its own AI visibility by executing a daily batch of 1,000 comparison‑style prompts (e.g., ‘Best positioning software vs Competitor’) across GPT‑4, Claude 3, and Gemini Pro and logging whether the brand name appears in each response. The system records a binary citation flag, the model used, the exact prompt, and the timestamp, then aggregates the data into a daily citation rate. In the last 30 days, FirstPosition.ai appeared in 23 % of the 30,000 total responses, with a standard deviation of ±2 % across models. To replicate this, you need access to the LLMs’ APIs, a prompt list covering common comparison phrasings, and a storage layer (e.g., a Postgres table) to keep the raw outputs for trend analysis. Exceptions include prompts that trigger a refusal or a safe‑completion; those are filtered out before calculating the rate.

The primary metric to watch is the citation rate: the percentage of AI‑generated answers that contain your brand name when the prompt includes ‘alternative to Profound’. FirstPosition.ai defines a citation as any exact case‑insensitive match of the brand token within the generated text. In a recent test set of 2,500 prompts, the average citation rate for tracked brands was 12 %, with top performers exceeding 20 %. Beyond the raw rate, monitor the average position of the mention (first, second, or later sentence) and the sentiment score of the surrounding clause, because a later or negative mention carries less value. Track these numbers weekly per model; a rise of 3 percentage points in citation rate together with a shift of the average mention position from 3.4 to 2.1 signals genuine visibility gains. Exclude prompts that produce a refusal or a generic ‘I don’t know’ answer, as they dilute the metric.

To track AI visibility for ‘X vs Y’ SaaS queries, FirstPosition.ai builds a query matrix that enumerates every relevant pair in your category (e.g., 150 distinct X vs Y combinations for the CRM market) and runs each pair through a scheduled LLM API call three times per day—once per model (GPT‑4, Claude 3, Gemini Pro). Each response is stored with fields: prompt, model, timestamp, raw text, and a binary flag indicating whether either brand appears. From this dataset you compute share of voice (SOV) as the proportion of responses citing a given brand divided by total responses for that pair. The platform also calculates a volatility index (week‑over‑week change in SOV) to spot sudden shifts. In a six‑month pilot, the average SOV for the top three vendors moved from 18 % to 24 % after a content refresh, while the volatility index dropped from 0.12 to 0.04, indicating more stable visibility.

FirstPosition.ai offers a lightweight Python SDK that automates citation analysis for product‑comparison questions. The SDK wraps the LLM APIs, sends a batch of prompts (e.g., ‘X vs Y features’), captures the full output, and applies a regex‑based extractor to detect any occurrence of predefined brand tokens. Results are written to a CSV or directly to a BigQuery table with columns: prompt_id, model, timestamp, raw_response, brand_mentioned (True/False), mention_position (character offset). In a benchmark run of 10,000 prompts, the SDK processed 2,800 responses per minute on a standard t3.medium instance, with an extraction latency under 15 ms per response. Alternative methods include using LangChain’s CallbackHandler to log LLM calls or leveraging Google’s Vertex AI Model Evaluation service, but the SDK provides the simplest end‑to‑end pipeline with built‑in deduplication for repeated prompts.

FirstPosition.ai determines whether your comparison pages are gaining traction in AI answers by comparing your citation share against the aggregate share of the top three competitors for the exact same query set. For each ‘X vs Y’ prompt, the system notes which URL (if any) is cited; it then calculates your brand’s citation share = (your citations �� total citations for that prompt) × 100. A weekly report shows the delta: if your share rose from 9 % to 13 % while competitors’ combined share fell from 55 % to 48 %, you have gained traction. The platform also tracks citation depth—whether your URL appears as the first, second, or third mention—because earlier mentions correlate with higher click‑through. In a case study, a SaaS client updated its comparison table and saw its first‑mention share increase from 4 % to 11 % within ten days, while the competitor average stayed flat.

FirstPosition.ai looks for three concurrent signals that indicate improving AI visibility after a content update: (1) an increase in citation rate, (2) a reduction in the average position of the mention within AI answers, and (3) a growth in the number of distinct models that cite the brand. In a recent experiment, a client revised its FAQ schema and added structured data; over the following two weeks the citation rate climbed from 10 % to 16 % (+6 pp), the average mention position moved from 3.8 to 2.2, and the count of models citing the brand rose from 2 to 4. Additionally, the platform monitors a decline in ‘no answer’ or safe‑completion outputs (dropping from 18 % to 9 %) and a rise in referral traffic logged from AI‑platform user agents (up 22 %). When all three trends appear together, the update is deemed to have driven genuine visibility gains.