AI agent for SEO and geo-SEO monument catalog

The challenge

The catalogue consisted of hundreds of pages of monuments, crosses and completed projects. Part of the cards contained only photos and technical fields, part of the — too short or repetitive texts. In the diagnosis of Yandex simultaneously appeared 32 pages with the same titles and 32 pages with the same descriptions.

It was necessary to make cards more useful for humans and understandable for search, but without mass reproduction of key phrases. Additional tasks related to geo-SEO for St. Petersburg and Leningrad region, internal linking, structured data, mobile version and correct operation of analytical cookie.

CMS is the source of the truth

The main limitation of the project is not to invent sizes, materials, equipment, price or terms. If the fact is not in MODX or it cannot be confirmed on the page, it does not appear in the text.

Existing fields were used for each card. price, price_all, time, material, decoration and desc. Photos were opened where it was necessary to understand the shape of the monument, the number of steles, the type of cross or the visible decoration elements. So the AI did not replace the data with guesses, but translated the already available information into user-readable text.

AI agent work cycle

The work was in small batches. This allowed you to change the style after feedback and not turn the entire catalog into a set of the same templates.

1. Search for blank documents

   The agent connected to MODX on SSH, selected the published goods of the desired section and found pages without description or FAQ.

   MODX Evolution · MySQL · sample from parent

2. Fact-finding

   Before writing, the title, the price of the monument and set, the material, design, production time, current text and available images were checked.

   price · price_all · time · material · decoration · desc

3. Description without SEO-perspam

   The text explained exactly what is shown in the photo, what is the difference between the project and what elements can be agreed individually. After checking the style, they abandoned the excessive grouping of characteristics so that the cards read naturally.

   Facts instead of blanks · different text structure and rhythm

4. Individual FAQ

   For the page, four to seven questions were created: about the cost composition, material, timing, inscriptions, portrait, delivery and installation. The wording was based only on the data of a specific card.

   4–7 questions · without non-existent installation prices

5. Publication and monitoring

   After saving, the cache was cleared, the HTTP response, titles, links, images and syntax of JSON-LD were checked. Changes were made with backup and control of specific documents.

   Backup · cache clear · HTTP 200 · JSON validation

What's changed in the cards

Geo-SEO without artificial city pages

Regional relevance was based on the real geography of the company. In the delivery and installation units, St. Petersburg and the Leningrad region were indicated, and in the descriptions of the completed projects, known installation sites were preserved. This information is useful for the customer, not a mechanical repetition of the name of the city.

Instead of doors for dozens of settlements, the main commercial pages, product cards and examples of work were strengthened. The search engine receives a bundle « service — product — material — region — executed project », and the user understands whether his area is served and whether it is possible to order an installation.

Expert content and linking

For the catalog prepared separate materials on the size of monuments, technical aspects of installation and forging. Then there were articles. forging material and about the use of forging in the composition of the monument.

Articles and cards are linked to service pages: forging, bronze and stone. References were made in meaning, not in the same block in all documents. The section of articles is translated to the conclusion of ten publications with pagination.

Technical SEO

• Unique metadata. Generation title and description Consider the card details, including material and price, when filled.
• Correction JSON-LD. Dynamic values are safely shielded, so quotation marks in cemeteries and other fields no longer break JSON.
• YML feed. A separate output of second-level documents for catalog data transmission has been prepared.
• Double control. Checked canonical, old addresses with ending .html and the single format URL.
• Check the results. After publication, meta tags, micromarkup, images and internal links were controlled directly to the HTML page.

Preparation for personal data requirements

A notification about cookie has been added to the site with a separate selection for the analytical cookie Yandex Metrics. The user can accept or reject the analytics, and the personal data processing policy is available from the notice. Consent is maintained so that the banner does not appear on each transition.

At the same time, adaptability was corrected: the correct viewportRestored styles after saving through the MODX manager, redesigned problematic mobile contact page blocks and limited content width on large screens.

Why this is not a mass AI generation

The AI agent did not receive the command « to write more text about the monuments ». He worked within a given process: one batch of documents, a fixed set of authorized sources, a ban on fictitious characteristics and a mandatory check of the published page.

When the wording looked unnatural — (for example, turned a simple colored inscription into a separate service for « registration variant of »), — it was rewritten. Feedback was used for subsequent batches rather than being masked by new keywords.

The result

The directory became more informative without increasing the text for the sake of keywords. Products have clear differences, answers to practical questions and links to services and expert articles. The technical part of the site received more stable metadata, valid markup and catalog feed, and the regional focus became visible from the real conditions of delivery, installation and work performed.

The main effect of the — approach is controllability. AI speeded up routine work with hundreds of documents, but the solutions remained verifiable: every fact can be traced back to the CMS field, a photo, or an existing page.

Open the monument-spb.ru → website