2026-09-18, GEO · Fundamentals · Rank Tracking

Extending an SEO program to AI answers: the migration map

An SEO program already has keyword lists, market splits, a tracking schedule, competitor sets and a reporting routine, and nearly all of it carries over to AI answers. The unit of measurement changes: a position in a list becomes a mention or a citation inside a written answer, and a single daily check becomes one sample from a distribution. Below, each SEO practice is mapped to its AI-answer equivalent, followed by a pipeline that runs both on one topic list and a list of what to drop.

The map: what carries over

The operating model carries over too. A keyword × market matrix becomes a prompt × market × engine matrix, and scheduled rank checks become async batches of up to 500 tasks, collected by polling or webhooks.

What changes shape

There is no single results page

Each engine searches, reads and cites differently. ChatGPT, Gemini, Perplexity, Copilot, Grok and AI Mode each need separate columns in every report. A blended “AI visibility score” hides the engine where you are absent, which is usually what you need to act on.

Results vary per run

A ranking checked twice in an hour usually returns the same position. An AI answer checked twice can name different brands and cite different pages. Treat each run as a draw: track rates over several runs, show the spread, and only call a change when it holds across runs. AI answer volatility and sampling vs census cover how many runs you need.

Answers are prose

A ranking is a position; an answer is written text. Your brand can be recommended, mentioned in passing, cited as evidence without being named, or absent. Mentions and citations are different events, and a report needs both.

Prompts are longer than keywords

“crm small business” becomes “What’s the best CRM for a 10-person sales team that already uses Gmail?” The added context (team size, existing tools, budget) changes which brands the answer recommends, so the prompt set needs several phrasings per intent, not one.

What stays the same

• Crawlability and indexing. Google’s AI features documentation says a page must be indexed and eligible to show in Search with a snippet to appear as a supporting link in AI Overviews or AI Mode, and that no extra requirements or special optimizations are needed (checked 2026-09-17).
• Snippet controls. The same page says nosnippet, data-nosnippet, max-snippet and noindex govern what Google’s AI features can show from a page. An aggressive snippet restriction added for SEO reasons also limits AI Overview and AI Mode use.
• Robots.txt, with new user agents. OpenAI’s crawler documentation separates OAI-SearchBot (sites opted out are not shown in ChatGPT search answers) from GPTBot (training; blocking it does not affect ChatGPT search results) (checked 2026-09-17). Audit robots.txt rules that were written before these agents existed.
• Content quality. Pages that answer the question directly, with specific and attributable facts, serve both rankings and citations.

Migrating a keyword program

1. Export the keyword list with its metadata: intent, market, priority, and the landing page you want to win.
2. Rewrite keywords as prompts. Keep one row per intent and write two or three natural phrasings. Add comparison (“X vs Y for Z”) and problem (“how do I…”) prompts; they are where assistants recommend products. Prompt set design has templates.
3. Map markets to request fields. Countries map to country. ChatGPT, Gemini, AI Mode and Google Search answer from 213 countries, and Copilot, Perplexity and Grok take any country; Gemini also takes a US state. Check each engine’s supported countries with GET /v1/countries?model=chatgpt (or gemini). For Google Search and AI Mode, keep your existing gl, hl and location settings.
4. Choose engines per segment. Not every prompt needs every engine. Commercial prompts for a consumer product might warrant both ChatGPT and Gemini; a B2B developer tool might start with ChatGPT alone.
5. Keep Google Search in the same run. Request the classic results with the AI Overview (include.aioverview) so you can compare ranking, AI Overview citation and assistant citation for the same topic.
6. Schedule and store. Submit weekly batches with idempotency keys derived from prompt, market, engine and date, and store every raw response. Idempotency keys explained shows the pattern.
7. Rebuild the report. Replace position charts with mention rate and citation share per engine, with competitor columns. An AI visibility report template is a starting layout.

Comparing rankings with citations

The most useful early report puts SEO and AI data side by side for each topic. For every keyword, compare the domains in Google’s organic top 10 with the domains cited in the AI Overview, in Gemini and in ChatGPT. Four cases come out:

The pipeline below, in Python with requests, submits one batch with three tasks per keyword, waits for results, and prints the comparison.

import os
import time
from urllib.parse import urlparse

import requests

API = "https://api.answerline.dev"
HEADERS = {"Authorization": f"Bearer {os.environ['ANSWERLINE_API_KEY']}"}
MY_DOMAIN = "acme.com"
DAY = "2026-09-20"

topics = [
    {"keyword": "invoicing software freelancers",
     "prompt": "What is the best invoicing software for freelancers?"},
    {"keyword": "time tracking app agencies",
     "prompt": "Which time tracking app works best for a small agency?"},
]

def host(url):
    h = urlparse(url).hostname or ""
    return h[4:] if h.startswith("www.") else h

tasks = []
for i, t in enumerate(topics):
    tasks += [
        {"taskType": "GOOGLE",
         "payload": {"query": t["keyword"], "country": "US", "include": {"aioverview": {"markdown": False}}},
         "idempotencyKey": f"bridge-google-{i}-US-{DAY}"},
        {"taskType": "GEMINI",
         "payload": {"prompt": t["prompt"], "country": "US"},
         "idempotencyKey": f"bridge-gemini-{i}-US-{DAY}"},
        {"taskType": "CHATGPT",
         "payload": {"prompt": t["prompt"], "country": "US"},
         "idempotencyKey": f"bridge-chatgpt-{i}-US-{DAY}"},
    ]

batch = requests.post(f"{API}/v1/async/task/batch", headers=HEADERS, json=tasks, timeout=360)
batch.raise_for_status()
ids = {r["index"]: r["task"]["id"] for r in batch.json()["results"] if r["success"]}

def wait(task_id):
    while True:
        body = requests.get(f"{API}/v1/async/task/{task_id}", headers=HEADERS, timeout=30).json()
        if body["task"]["status"] in ("COMPLETED", "FAILED"):
            return body
        time.sleep(10)

for i, t in enumerate(topics):
    if any(i * 3 + k not in ids for k in range(3)):
        print(t["keyword"], "has a task that was not created; see the batch results")
        continue
    google, gemini, chatgpt = (wait(ids[i * 3 + k]) for k in range(3))
    if "FAILED" in (google["task"]["status"], gemini["task"]["status"], chatgpt["task"]["status"]):
        print(t["keyword"], "has a failed task; resubmit later")
        continue
    g, a, c = google["response"]["result"], gemini["response"]["result"], chatgpt["response"]["result"]

    organic = {host(r["link"]) for r in g.get("organicResults", [])[:10]}
    overview = g.get("aioverview") or {}
    cited = {
        "ai_overview": {host(s["url"]) for s in overview.get("sources", [])},
        "gemini": {host(s["url"]) for s in a.get("sources", [])},
        "chatgpt": {host(s["url"]) for s in c.get("sources", [])},
    }
    print(t["keyword"])
    print("  ranks top 10:", MY_DOMAIN in organic)
    for engine, domains in cited.items():
        print(f"  cited in {engine}:", MY_DOMAIN in domains, "| cited but not ranking:", sorted(domains - organic)[:5])

The polling loop keeps the example short. For hundreds of keywords, add a webhook to each task instead and let results arrive; see rank tracking pipelines with batches.

Fan-out coverage

Fan-out coverage, the share of an engine’s own searches you rank for, needs those searches. ChatGPT’s searchQueries[] (with include.searchQueries, which is free) comes back empty on the answers served today, so the check cannot be run for ChatGPT. The “cited but not ranking” domains above are the link between the two programs instead. Query fan-out covers the idea.

Worked cost example

For example, 200 keywords run weekly is $0.588 per run and $2.35 over four runs. The tiers, and how 30-day spend lowers the price, are on the pricing page; cost planning goes further.

What to drop

• Blended visibility scores that average engines into one number with no way back to the underlying answers.
• Daily position charts without variance. For answers, a single daily point overstates precision; show rates across runs.
• Metrics that cannot be traced to a stored response. If a client asks “where did we lose that mention?”, you should be able to open the exact answer text and sources.
• Branded-only tracking. “What is Acme?” is the least informative prompt; category and comparison prompts decide consideration.

Pitfalls during the migration

• Translating keywords word for word. Prompts need natural phrasing and context, not keyword syntax.
• Changing the prompt set mid-trend. Version the set; add prompts as a new cohort so historical rates stay comparable.
• Comparing engines on different days. Run all engines for a topic in the same batch so differences are not timing artifacts.
• Ignoring failed tasks. A FAILED task is not charged; resubmit it rather than counting it as “not mentioned”.

Start with the quickstart, or see rank tracking and AI visibility tracking for the two halves of the hybrid program.

Keep reading