AM Many searches follow a pattern: "emergency plumber in Austin", "emergency plumber in Denver", "Klein ET120 vs Fluke 1630", "best laptop for video editing". Programmatic SEO answers a pattern like that with one template and a table of data. Each row becomes its own page, aimed at one specific search.
The hard part is not making the pages. It is making each page worth reading. Programmatic SEO in AM Jarvis is built around that problem: it scores whether a niche deserves this approach at all, asks you to verify the data, and checks every finished page against its siblings before it lets Google index it.
How a build works
- Score the niche. Enter your niche, market and business model (affiliate, lead gen, tools, directory, ecommerce or ads) and click Score this niche.
- Build the dataset. Pick a page family and a pattern, list your fields, then paste your own rows as CSV or click Generate for AI starter rows.
- Check the data. Verify every value and fill in source_url and verified_at for each row.
- Set the template. Name the project, set an optional title pattern, pick the Unique Value Gate threshold, theme, depth and language, then click Build programmatic site.
- Download or publish. When the project shows ready, download it from Projects or publish it to your hosting.
Step 1: the niche score
The score rates your niche from 0 to 10 on 8 factors: repeatable query patterns, available structured data, distinct intent between page variants, room for unique value per page, commercial value, how easy the data is to keep current, internal linking potential and SERP opportunity. The average times 10 gives a score from 0 to 100.
| Score | Verdict |
|---|---|
| 80 and up | Go |
| 70 to 79 | Pilot first |
| 60 to 69 | Only with a moat (data or a calculation others do not have) |
| Below 60 | Redesign the idea |
You also get page patterns you can tap to use, a list of combinations that should not become pages, the minimum fields your dataset needs, a suggested 50-page pilot and the main policy risks. If you have not settled on a niche yet, Niche Research shows what people actually type into Google first.
Step 2: the page family
The family decides what makes each page different and which sections the AI writes.
| Family | One page per | Example pattern | Schema |
|---|---|---|---|
| Entity | Product, item or record | [name] Review: Specs, Price and Who It's For | Product |
| Service by location | City or area | [service] in [city] | Service |
| Comparison | A vs B pair | [a] vs [b] | Article |
| Best-for | Use case | best [name] for [use_case] | ItemList |
Words in [brackets] are placeholders that must match your column names. A row where service is Plumbing and city is Austin becomes "Plumbing in Austin".
What a good dataset looks like
Your dataset decides whether the pages are useful. The template and the AI cannot add facts you do not have, and they are told not to try.
- Facts that change the answer. A city page needs local numbers that differ between cities: callout fees, response times, permit rules, opening hours. If every row has the same values, every page will say the same thing.
- A source for each row. Fill source_url with an official or reputable page, and verified_at with the date you checked it. Both show on the page under the data table.
- Few empty cells. A row needs at least 40% of its fields filled to be built.
- Only searches people make. Skip combinations nobody looks for, or where the answer does not change with the variable.
The Generate button can draft 10 to 40 starter rows. Treat them as a to-do list, not facts: the AI is asked for real entities, but it can still be wrong, and it leaves verified_at empty on purpose. A build needs at least 3 records and takes up to 500.
What goes on each page
Every page opens with a data table rendered straight from your row, so the numbers are exactly what you entered, followed by the source link. The AI then writes only the page-specific sections from that row's data. For a location page those are a quick answer, local details, how to choose in that area, links to nearby area pages and three location-specific questions. Standard depth writes 550 to 800 words per page; Deep writes 900 to 1,200. Pages can be written in English, Spanish, German, French, Italian, Portuguese, Dutch, Arabic or Urdu.
Each page links to up to 3 related siblings, chosen from rows that share a category value where possible. A hub page lists every page, with indexable pages first, so the site forms a hub, entity and related page graph instead of a pile of orphan pages.
The Unique Value Gate
Google's spam policies name scaled content abuse: many pages generated mainly to manipulate rankings rather than help people, including AI-generated pages that add no value. The same page lists doorway abuse, such as many pages aimed at different cities that funnel visitors to one place. A template site with the city name swapped and nothing else is exactly what those policies describe.
The Gate is how AM Jarvis keeps a build on the right side of that line. After every page is written, it removes each page's own data values (the city name, the prices) from the text, then compares the remaining three-word phrases with every sibling page. A page's score is its overlap with its closest sibling. Swapping the city name alone never makes a page unique, because the city name is taken out before the comparison.
| Setting | Noindex when a page shares more than | Use it when |
|---|---|---|
| Strict | 30% of its wording with a sibling | You want the safest result and accept fewer indexed pages |
| Standard (default) | 40% | Most sites |
| Relaxed | 50% | Rows are rich and you accept more thin content risk |
Pages that fail get a noindex,follow robots tag. They are still published and linked, so visitors can open them and Google can still follow their links, but Google is asked not to list them and they stay out of the sitemap. Rows missing their identifying field, or with too many empty fields, are held and not built at all. Pages the AI failed to write are noindexed too.
Warning: The Gate measures wording overlap. It cannot check whether your facts are true or whether anyone searches for a page. Passing the Gate is not a promise that Google will index or rank a page.
Worked example: 50 city pages
Say you run the pattern "[service] in [city]" for emergency plumbing with 50 rows and 9 columns: city, state, licensed_plumbers, avg_callout_fee, after_hours_available, typical_response_time, permit_rule, source_url and verified_at. The numbers below show how the rules play out.
- Held rows. With 9 columns, a row needs 4 filled values. Three small towns have only city, state and source_url, so they are held. 47 pages get built.
- Gate at Standard (40%). Six suburbs of one metro area share the same fee, response time and permit rule, so their pages read almost the same. Each overlaps its closest sibling by more than 40% and gets noindex. 41 pages pass.
- Sitemap. sitemap-pseo.xml lists the 41 passing pages plus the hub, About and Contact pages. The hub shows 47 pages built from 50 records, 41 indexable.
- Fix and rebuild. You find a real difference for the six suburbs, such as which ones require a permit for a water heater swap, and fill in the three thin towns. The next build can pass more pages.
What you get in the download
- One HTML page per built row, each with a self-referencing canonical tag and JSON-LD schema for its family
- A hub homepage with CollectionPage schema, plus About and Methodology, Contact, Privacy Policy and Terms pages
- sitemap.xml pointing to sitemap-pseo.xml, which lists only indexable pages
- robots.txt that allows crawling and points to the sitemap
- dataset.csv, the exact rows used
- VALUE-GATE-REPORT.txt, with every page's status, its similarity score, its closest sibling and the reason each held row was skipped
The files use YOURDOMAIN.com as a placeholder. The Contact page also carries a placeholder email, so replace it with your real address. If you need a different robots.txt, the robots.txt generator builds one. Location pages carry Service schema; a real business with an address can add LocalBusiness markup from the schema markup generator.
Publish to your hosting
Finished sites also appear in the Site Builder table. From Publish Sites you can upload one straight to your hosting with FTPS, FTP or a cPanel API token. If you set the site address on the hosting account, the https://YOURDOMAIN.com placeholder is replaced with it during upload. You can then send the sitemap URLs to IndexNow, and since noindexed pages are not in the sitemap, they are not submitted. For hand-written niche sites instead of data-driven ones, use the AI Site Builder.
What it does not do
- It does not verify your data. AI starter rows can be wrong until you check them.
- It does not invent local facts, reviews or test results. If a value is missing, the page says less.
- It does not guarantee indexing or rankings.
- It does not edit a live site. To change pages, update the dataset and build again as a new project.
- It does not upload to your hosting on its own. You choose when to publish.
Troubleshooting
Many pages are noindexed
Open VALUE-GATE-REPORT.txt and look at which pages are twins. The fix is almost always the data: the rows are too alike, so the pages are too. Add a column that genuinely differs per row, fill empty cells, or drop rows that only exist to target a keyword. Moving to Relaxed hides the symptom and raises the thin content risk.
Rows were held
The report names each held row and the reason: a missing identifying field, or too many empty fields (for example 3 of 9). Fill in real values and build again. If you cannot find the facts, that row probably should not be a page.
A page shows almost no text
If the AI call fails or returns under 150 words, the page gets a short placeholder and noindex. Check your AI settings and build again.
Plan availability
The Programmatic SEO engine is included in the Agency plan only, with no monthly cap on projects. See the pricing page for current prices. Once pages are live, add a sample of target searches to Rank Tracker to see which ones gain ground.